{ "text": "大家好 欢迎收看新一期视频\nDeepseek刚刚发布了一个很有意思的更新\nDeepseek V4 Flash官方API现已开启公测\n新版本命名为Deepseek V4 Flash 0731\n这可不只是一个小补丁\nDeepseek表示他们大幅提升了智能体能力\n在该模型上甚至\n使其测试分数如今已远超V4 Pro预览版\n也就是说\n这个小模型现在击败了预览版\n极大模型在智能体任务上的版本\n这确实挺疯狂的\n我也用我自己的Kinbench测试了它\n而在我的基准测试中\n它做到了\n其他任何模型都未曾做到的事\n包括Code 3.5\n以及Opus稍后我们会细聊\n但首先\n我们先聊聊这次更新到底是什么\n前阵子我介绍过DeepSeek V4 Pro\n和V4 Flash的预览版\n包括如何通过NVIDIA Link免费使用它们\n视频链接已放在下方\n我们先快速回顾一下Flash\n它是V4系列中较小的模型\n该混合专家模型拥有2000\n0840亿的总参数量\n约130亿激活参数\n支持100万Token上下文窗口\n它定位为主打快速\n与低成本版本\n而V4 Pro\n是1.6万亿参数旗舰大模型\n关于这次0731更新\n有一个关键细节\nDeepSea明确表示\nV4Flash 0731\n保持了完全相同的模型架构\n和参数规模与预览版完全一致\n所以它并没有变大\n依然是\n那款2840亿参数的Flash\n但在此基础上\n它似乎进行了高强度的智能体后训练\n所有的性能提升都源于训练的优化\n而不是增加参数量\n他们还明确指出\n这次升级仅是用于DeepSeek V4 Flash API\nV4 Pro API\n以及App端D模型\n和网页端目前都保持不变\n公告最后提到\nDeepSeek V4 Pro的正式版本\n很快就会发布\n所以这次Flash更新基本只是热身\n真正的重头戏还在后头\n另外\n这次发布中还有一点\n我觉得非常明智\n官方V4Flash现在原生支持Response API格式\nDipsy还表示他已完全适配Codex\n所以他们不仅是发布了一个模型\n而不是被动等待工具链的适配\n他们直接面向Codex适配框架进行了开发\n这意味着你可以将其接入\n到Codex风格的工作流中且毫无阻碍\nGLM以及国内的其他一些实验室也在做\n类似的框架适配\n事实证明效果立竿见影\n这确实是个明智之举\n接下来看看官方的基准测试数据\n这次性能提升太惊人了\n在Terminal Bench 2.1测试中\n新版Flash跑分高达82.7\n而此前的预览版仅为61.8\n甚至连V4 Pro预览版也只有72.1分\n这也让它超越了81分的GLM40520\n而且距离85分的OPUS差距并不大\n要知道后者的价格可要贵得多\n而最夸张的是SWBEC\n预览版在这个测试中仅获得了7.3分\n而新版本直接飙升到54.4分\n实现了从基本无法使用\n到足以于46.2分的GLM40520正面竞争的水平\n甚至是58分的OPUS\nCybergen也从38.7分\n升至76.7分涂了冷验证机\n从49.7分升至70.3分\n在其内部DSBench测试集上\n全战任务得分从37升至68.7\n高难度编码智能体任务也\n从25.8升至59.6\n总的来说\n这叫其自身的预览板有了巨大飞跃\n且在大多数智能体基准测试中\n它已逼近OPUS 4.8\n而它作为一款尺寸小得多\n且更便宜的模型\n不过需要说明的是\nDeep-C指出\n在公开的编码智能体任务测试中\nV4 Flash使用了\n其即将推出的Deep-C Harness框架在极简模式下\n以最高档位运行\n且套采样设为0.95\n温度参数设为一可见\n测试框架的选择至关重要\n测试结果可能不同\n数据仅供参考\n不过了解本频道的人都知道\n我们从不只看官方宣传数据\n所以我用自己的基准测试跑了一下\n话不多说我们直接切入正题\n和往常一样\n我用我的TingBench进行了测试\n其中涵盖了一些前端\n与动画一些FreeJS任务以及SVG测试\n一道数学题以及一个长城智能体任务\n最后还有一个超难的3D任务\n每项测试满分10分\n最后我们会通过最终的图表来看看\n看看它的表现与其他模型相比如何\n第一题是电梯模拟\n模型需要构建一个模拟程序\n你可以在不同楼层生成乘客\n共有三部电梯\n其每部只能在一人\n没赶上的人需要搭乘下一班电梯\n还需要有提示框显示每个人的目标楼层\n当鼠标悬停在他们身上时\n我们发送提示词看看效果\n这个表现只能算一半\n得了五分基本框架是有了\n但关于乘客没赶上电梯\n并被下一班电梯接走的核心逻辑并没有完全跑通\n而且动画也不够流畅\n作为参考\n只有OPUS模型在这项测试中拿过10分\n而像CLUD3.5这样的模型\n和KIMMYK3也只有9分\n所以这开局只能算重归重矩\n第二个测试是FreeJF隐形眼镜盒\n它需要生成一个3D模型\n是一个带有明显L和R标识盖子的隐形眼镜盒\n并且点击盖子能够将其打开这项测试\n几乎所有模型都会在这里翻车\n我们发送过去看看\n好的 这其实做得非常好\n它拿到了8分\n盒子看起来很像样\n像个合格的眼镜盒L\n而且L和二盖也清晰可见\n点击打开的交互也有效\n这让他与Quan2.5Maps并列\n而Opus拿到了5分\n是历史第二高分\n在该测试中仅次于Plus 3.5\n他仍然是唯一在这个测试中\n拿到满分10分的模型\n所以对于一个小模型来说\n能有这样的表现已经非常惊艳了\n第三题是\nFreeJF折叠桌\n这里有一个滑块\n当你向右拖动它时\n桌子应该展开\n向左拖则折叠回去\n动画需要无缝衔剑\n且呈现出真正的3D效果\n我们发送请求看看\n这个得到了7分\n桌子确实能折叠起来\n并能随滑快拖动而展开\n过度基本自然\n只是还不够完全流畅\n表现最好的模型\n如Cloud 3.5\nKIMI GLM4\n以及Summit 3.5\n在这项测试中都拿到了9分\n但对比来看\nOpus在这项测试中仅得了5分\n所以这表现绝对算很不错了\n第四题是生成熊猫的SVG\n吃着汉堡\n我们发送过去看看\n好\n这确实能看出是一只熊猫\n也能看出是汉堡\n但构图有点别扭\n而且看起来不太像在吃汉堡\n这题给五分\n在这项测试中\n像Pimi和Gemini 1.5这样表现最好的模型\n以及Quain 2.5 Max得分也仅为8分\n所以并不算惊艳\n第五题是制作弓箭\n模拟游戏里面设计了四个靶子\n玩家需要以最短的时间击中所有靶子\n从而登上排行榜\n我们发送代码运行一下看看\n这道题得了7分\n核心射击机制可以正常运行\n靶子能判定击中\n排行榜功能也正常\n但整体手感和细节打磨\n还达不到顶尖模型的水平\n比如Grog 1.5和Quang 2.5 Math\n虽然Opus模型在这里都拿了满分\n但这依然是个可玩的游戏\n所以表现还算不错\n第六题是一道数学难题\n关于计算有序队的排列数\n答应该是2460\n很多模型在这一题上完全翻车\n我们提交一下\n看看\n他完美答对了2460\n完全正确\n拿了满分10分\n这让他机身同一梯队\n与Cloud 3.5\nTimmy Chat\nQueen Max\n以及Opus模型并列\n有趣的是\n在我测试时\nV4 Pro预览版也做对了这道题\n所以V4系列的推理能力显然非常过瘾\n第七题是一个复杂的长城任务\n生成一个关于熊猫知识的数据集\n并用该数据集为调一个GEM2B模型\n接着构建一个本地WebUI每次刷新页面都会生成一条熊猫冷知识\n全程从零开始完全自主完成\n且完全在本地运行\n我们启动任务看看效果\n这次的表现绝对可以达满分\n它生成了数据集并完成了整个微调\n独自运行期间完全没有卡科\n生成的Web UI正常运行\n每次刷新都显示新的熊猫冷知识\n这正是那种复杂的长程智能体任务\nDeep C声称对其进行过针对性升级\n说实话\n效果显而易见\n之前的预览版远没这么可靠\n这波表现确实堪称完美\n第八题也是最后一题\n是制作一个3D万秒\n要求是一个功能完整的3D万秒\n能显示当前实际时间\n秒针 分针\n以及时针都要平滑转动\n还要有日期\n核心器显示\n以及双时区\n这是整个测试集中最难的一题\n之前没有任何模型\n能在这道题上拿到好成绩\n历史最高分是Fibo5创下的\n4分\n让我们直接运行测试\n但你感谢\nLip6 V4 Flav竟然拿到了6分\n创下这道题历史记录\n手表完整呈现出来\n指针转动平滑\n只能实时同步时间\n双时区设置也基本正常\n虽然还不完美\n所以没能拿到更高的分数\n但这个小模型刚刚击败了\nCloud 3.5 Opus\n以及我测试集最难题目上的\n其他所有前沿模型\n这完全出乎我的意料\n来看一下最终的图表\nDeep Seek V4 Flash 0731\n最终拿到了58分\n得分率为72.5%\n这让它排在\n略高于GPT4O513的71.25%\n稍低于Gemini 1.5 Pro的75%\n而在它之上\n有Kinning和\nOpas并列77.5%\nGTT 450808达80%\n接着是Quain\nQuain 2.5 Max拿下81.25%\n而Cloud 3.5仍以82.5%稳居榜首\n而对比以下背景\n你就能明白这有多惊人\n我之前测试Deep Seek V4 Pro预览版时\n它在此测试中仅得了24.8%\n新的Flash模型得分达到了72.5%\n这意味着该系列的小模型现在\n比大模型预览版的表现提升了近三倍\n在我的测试中\n而且在架构和尺寸与预览版相同\n单靠后训练就能取得惊人提升\n不可思议\n这也印证了Deep Seed所宣称的Agent能力升级\n从得分规律中也可以清楚地看到\n所有射击推理\n或长城Agent任务基本上都拿到了满分\n比如数学和微调任务\n还有创纪录的时钟生成常识\n但像熊猫SVG这种纯视觉打磨的任务\n还有电梯动画\n在这些方面它的表现依然平平\n所以这确实很像是一个\n针对智能体编程深度微调的模型\n正如官方基准测试所示\n那么结论是什么\n这绝对是目前性价比最高的模型之一\n可用于智能体编程\n在我的测试中\n它的表现已经达到了Gemini 1.5 Pro的水平\n它可以直接接入Cursor等代码工具\n这得益于其API知识\n而且Ditsig的价格\n一直都非常便宜\n如果你是学生\n或者对于预算有限的开发者\n这绝对是个极佳的选择\n他显然还无法击败Cloud 3.5\n或是在综合表现上超越OPUS\n而且他的视觉能力依然表现平平\n但考虑到他的体量和价格\n这已经足够经验了\n而且别忘了\n这还仅仅是Flash版本\n官方的DeepSick V4 Pro也即将发布\n据官方介绍\n如果将同样的智能体后训练应用\n到1.6万亿参数的模型上\n那对市面上昂贵的B元模型来说\n将是降维打击般的存在\n我会在上线后第一时间进行测试\n尽情关注\n总的来说\n这非常酷", "language": null, "duration": 666.900313, "segments": [ { "start": 0.0, "end": 7.46, "text": "大家好 欢迎收看新一期视频", "chunk": 0, "language": "zh", "avg_logprob": -0.13818857691309475, "no_speech_prob": 1.065416678491804e-11, "compression_ratio": 1.1286307053941909, "source": "whisper" }, { "start": 7.46, "end": 10.72, "text": "Deepseek刚刚发布了一个很有意思的更新", "chunk": 0, "language": "zh", "avg_logprob": -0.13818857691309475, "no_speech_prob": 1.065416678491804e-11, "compression_ratio": 1.1286307053941909, "source": "whisper" }, { "start": 10.72, "end": 14.98, "text": "Deepseek V4 Flash官方API现已开启公测", "chunk": 0, "language": "zh", "avg_logprob": -0.13818857691309475, "no_speech_prob": 1.065416678491804e-11, "compression_ratio": 1.1286307053941909, "source": "whisper" }, { "start": 14.98, "end": 19.24, "text": "新版本命名为Deepseek V4 Flash 0731", "chunk": 0, "language": "zh", "avg_logprob": -0.13818857691309475, "no_speech_prob": 1.065416678491804e-11, "compression_ratio": 1.1286307053941909, "source": "whisper" }, { "start": 19.24, "end": 21.6, "text": "这可不只是一个小补丁", "chunk": 0, "language": "zh", "avg_logprob": -0.13818857691309475, "no_speech_prob": 1.065416678491804e-11, "compression_ratio": 1.1286307053941909, "source": "whisper" }, { "start": 21.6, "end": 25.0, "text": "Deepseek表示他们大幅提升了智能体能力", "chunk": 0, "language": "zh", "avg_logprob": -0.13818857691309475, "no_speech_prob": 1.065416678491804e-11, "compression_ratio": 1.1286307053941909, "source": "whisper" }, { "start": 25.0, "end": 27.18, "text": "在该模型上甚至", "chunk": 0, "language": "zh", "avg_logprob": -0.13818857691309475, "no_speech_prob": 1.065416678491804e-11, "compression_ratio": 1.1286307053941909, "source": "whisper" }, { "start": 27.18, "end": 30.6, "text": "使其测试分数如今已远超V4 Pro预览版", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 30.6, "end": 31.78, "text": "也就是说", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 31.78, "end": 33.98, "text": "这个小模型现在击败了预览版", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 33.98, "end": 36.72, "text": "极大模型在智能体任务上的版本", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 36.72, "end": 38.42, "text": "这确实挺疯狂的", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 38.42, "end": 40.42, "text": "我也用我自己的Kinbench测试了它", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 40.42, "end": 42.32, "text": "而在我的基准测试中", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 42.32, "end": 43.04, "text": "它做到了", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 43.04, "end": 45.2, "text": "其他任何模型都未曾做到的事", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 45.2, "end": 46.4, "text": "包括Code 3.5", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 46.4, "end": 48.36, "text": "以及Opus稍后我们会细聊", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 48.36, "end": 48.86, "text": "但首先", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 48.86, "end": 51.04, "text": "我们先聊聊这次更新到底是什么", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 51.04, "end": 53.34, "text": "前阵子我介绍过DeepSeek V4 Pro", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 53.34, "end": 55.12, "text": "和V4 Flash的预览版", "chunk": 0, "language": "zh", "avg_logprob": -0.14506800320683694, "no_speech_prob": 1.5655397117564362e-11, "compression_ratio": 1.2421652421652423, "source": "whisper" }, { "start": 55.12, "end": 58.12, "text": "包括如何通过NVIDIA Link免费使用它们", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 58.12, "end": 59.56, "text": "视频链接已放在下方", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 59.56, "end": 60.94, "text": "我们先快速回顾一下Flash", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 60.94, "end": 63.12, "text": "它是V4系列中较小的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 63.12, "end": 65.4, "text": "该混合专家模型拥有2000", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 65.4, "end": 67.72, "text": "0840亿的总参数量", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 67.72, "end": 69.66, "text": "约130亿激活参数", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 69.66, "end": 72.0, "text": "支持100万Token上下文窗口", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 72.0, "end": 73.78, "text": "它定位为主打快速", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 73.78, "end": 75.18, "text": "与低成本版本", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 75.18, "end": 75.9, "text": "而V4 Pro", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 75.9, "end": 78.2, "text": "是1.6万亿参数旗舰大模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 78.2, "end": 79.94, "text": "关于这次0731更新", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 79.94, "end": 81.04, "text": "有一个关键细节", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 81.04, "end": 82.5, "text": "DeepSea明确表示", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 82.5, "end": 84.34, "text": "V4Flash 0731", "chunk": 0, "language": "zh", "avg_logprob": -0.11961187705015525, "no_speech_prob": 1.7848923727914467e-11, "compression_ratio": 1.151862464183381, "source": "whisper" }, { "start": 84.34, "end": 86.54, "text": "保持了完全相同的模型架构", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 86.54, "end": 89.06, "text": "和参数规模与预览版完全一致", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 89.06, "end": 90.7, "text": "所以它并没有变大", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 90.7, "end": 91.3, "text": "依然是", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 91.3, "end": 93.48, "text": "那款2840亿参数的Flash", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 93.48, "end": 94.82, "text": "但在此基础上", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 94.82, "end": 96.88, "text": "它似乎进行了高强度的智能体后训练", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 96.88, "end": 99.64, "text": "所有的性能提升都源于训练的优化", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 99.64, "end": 101.2, "text": "而不是增加参数量", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 101.2, "end": 102.24, "text": "他们还明确指出", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 102.24, "end": 105.04, "text": "这次升级仅是用于DeepSeek V4 Flash API", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 105.04, "end": 106.48, "text": "V4 Pro API", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 106.48, "end": 108.24, "text": "以及App端D模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 108.24, "end": 110.12, "text": "和网页端目前都保持不变", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 110.12, "end": 112.0, "text": "公告最后提到", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 112.0, "end": 114.24, "text": "DeepSeek V4 Pro的正式版本", "chunk": 0, "language": "zh", "avg_logprob": -0.11138702681844269, "no_speech_prob": 1.5541590583922904e-11, "compression_ratio": 1.2458100558659218, "source": "whisper" }, { "start": 114.24, "end": 115.96, "text": "很快就会发布", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 115.96, "end": 118.24, "text": "所以这次Flash更新基本只是热身", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 118.24, "end": 120.78, "text": "真正的重头戏还在后头", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 120.78, "end": 121.72, "text": "另外", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 121.72, "end": 123.26, "text": "这次发布中还有一点", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 123.26, "end": 125.02, "text": "我觉得非常明智", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 125.02, "end": 128.94, "text": "官方V4Flash现在原生支持Response API格式", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 128.94, "end": 131.62, "text": "Dipsy还表示他已完全适配Codex", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 131.62, "end": 133.82, "text": "所以他们不仅是发布了一个模型", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 133.82, "end": 135.88, "text": "而不是被动等待工具链的适配", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 135.88, "end": 138.28, "text": "他们直接面向Codex适配框架进行了开发", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 138.28, "end": 140.32, "text": "这意味着你可以将其接入", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 140.32, "end": 142.98, "text": "到Codex风格的工作流中且毫无阻碍", "chunk": 0, "language": "zh", "avg_logprob": -0.1084110612814137, "no_speech_prob": 1.4544548725126116e-11, "compression_ratio": 1.2275280898876404, "source": "whisper" }, { "start": 142.98, "end": 145.74, "text": "GLM以及国内的其他一些实验室也在做", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 145.74, "end": 146.94, "text": "类似的框架适配", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 146.94, "end": 148.84, "text": "事实证明效果立竿见影", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 148.84, "end": 150.1, "text": "这确实是个明智之举", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 150.1, "end": 152.44, "text": "接下来看看官方的基准测试数据", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 152.44, "end": 153.82, "text": "这次性能提升太惊人了", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 153.82, "end": 155.18, "text": "在Terminal Bench 2.1测试中", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 155.18, "end": 157.44, "text": "新版Flash跑分高达82.7", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 157.44, "end": 159.96, "text": "而此前的预览版仅为61.8", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 159.96, "end": 162.98, "text": "甚至连V4 Pro预览版也只有72.1分", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 162.98, "end": 166.26, "text": "这也让它超越了81分的GLM40520", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 166.26, "end": 169.16, "text": "而且距离85分的OPUS差距并不大", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 169.16, "end": 171.3, "text": "要知道后者的价格可要贵得多", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 171.3, "end": 172.8, "text": "而最夸张的是SWBEC", "chunk": 0, "language": "zh", "avg_logprob": -0.09934688144259983, "no_speech_prob": 1.7365397314561548e-11, "compression_ratio": 1.1428571428571428, "source": "whisper" }, { "start": 172.98, "end": 175.58, "text": "预览版在这个测试中仅获得了7.3分", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 175.58, "end": 178.04, "text": "而新版本直接飙升到54.4分", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 178.04, "end": 179.9, "text": "实现了从基本无法使用", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 179.9, "end": 183.9, "text": "到足以于46.2分的GLM40520正面竞争的水平", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 183.9, "end": 185.92, "text": "甚至是58分的OPUS", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 185.92, "end": 188.14, "text": "Cybergen也从38.7分", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 188.14, "end": 190.68, "text": "升至76.7分涂了冷验证机", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 190.68, "end": 193.28, "text": "从49.7分升至70.3分", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 193.28, "end": 195.5, "text": "在其内部DSBench测试集上", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 195.5, "end": 198.26, "text": "全战任务得分从37升至68.7", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 198.26, "end": 200.38, "text": "高难度编码智能体任务也", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 200.38, "end": 202.74, "text": "从25.8升至59.6", "chunk": 0, "language": "zh", "avg_logprob": -0.14357845366947233, "no_speech_prob": 1.4006852969150607e-11, "compression_ratio": 1.1935483870967742, "source": "whisper" }, { "start": 202.74, "end": 203.8, "text": "总的来说", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 203.8, "end": 205.84, "text": "这叫其自身的预览板有了巨大飞跃", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 205.84, "end": 208.2, "text": "且在大多数智能体基准测试中", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 208.2, "end": 209.82, "text": "它已逼近OPUS 4.8", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 209.82, "end": 211.82, "text": "而它作为一款尺寸小得多", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 211.82, "end": 213.16, "text": "且更便宜的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 213.16, "end": 214.1, "text": "不过需要说明的是", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 214.1, "end": 215.12, "text": "Deep-C指出", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 215.12, "end": 216.82, "text": "在公开的编码智能体任务测试中", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 216.82, "end": 218.38, "text": "V4 Flash使用了", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 218.38, "end": 221.36, "text": "其即将推出的Deep-C Harness框架在极简模式下", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 221.36, "end": 222.76, "text": "以最高档位运行", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 222.76, "end": 224.12, "text": "且套采样设为0.95", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 224.12, "end": 225.84, "text": "温度参数设为一可见", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 225.84, "end": 227.5, "text": "测试框架的选择至关重要", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 227.5, "end": 229.0, "text": "测试结果可能不同", "chunk": 0, "language": "zh", "avg_logprob": -0.11585808648003472, "no_speech_prob": 1.5326632324397238e-11, "compression_ratio": 1.2240437158469946, "source": "whisper" }, { "start": 229.0, "end": 230.58, "text": "数据仅供参考", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 230.58, "end": 232.66, "text": "不过了解本频道的人都知道", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 232.66, "end": 234.64, "text": "我们从不只看官方宣传数据", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 234.64, "end": 236.38, "text": "所以我用自己的基准测试跑了一下", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 236.38, "end": 238.36, "text": "话不多说我们直接切入正题", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 238.36, "end": 239.44, "text": "和往常一样", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 239.44, "end": 240.86, "text": "我用我的TingBench进行了测试", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 240.86, "end": 242.4, "text": "其中涵盖了一些前端", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 242.4, "end": 245.2, "text": "与动画一些FreeJS任务以及SVG测试", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 245.2, "end": 248.78, "text": "一道数学题以及一个长城智能体任务", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 248.78, "end": 250.72, "text": "最后还有一个超难的3D任务", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 250.72, "end": 252.5, "text": "每项测试满分10分", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 252.5, "end": 255.28, "text": "最后我们会通过最终的图表来看看", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 255.28, "end": 258.06, "text": "看看它的表现与其他模型相比如何", "chunk": 0, "language": "zh", "avg_logprob": -0.12722485673193837, "no_speech_prob": 1.7504652241595586e-11, "compression_ratio": 1.326975476839237, "source": "whisper" }, { "start": 258.06, "end": 259.88, "text": "第一题是电梯模拟", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 259.88, "end": 262.3, "text": "模型需要构建一个模拟程序", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 262.3, "end": 264.52, "text": "你可以在不同楼层生成乘客", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 264.52, "end": 266.02, "text": "共有三部电梯", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 266.02, "end": 267.5, "text": "其每部只能在一人", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 267.5, "end": 269.98, "text": "没赶上的人需要搭乘下一班电梯", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 269.98, "end": 272.76, "text": "还需要有提示框显示每个人的目标楼层", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 272.76, "end": 274.72, "text": "当鼠标悬停在他们身上时", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 274.72, "end": 276.5, "text": "我们发送提示词看看效果", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 276.5, "end": 278.02, "text": "这个表现只能算一半", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 278.02, "end": 280.0, "text": "得了五分基本框架是有了", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 280.0, "end": 282.02, "text": "但关于乘客没赶上电梯", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 282.02, "end": 284.62, "text": "并被下一班电梯接走的核心逻辑并没有完全跑通", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 284.62, "end": 286.5, "text": "而且动画也不够流畅", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 286.5, "end": 287.36, "text": "作为参考", "chunk": 0, "language": "zh", "avg_logprob": -0.09506281366888082, "no_speech_prob": 1.68271663963937e-11, "compression_ratio": 1.4298245614035088, "source": "whisper" }, { "start": 287.36, "end": 289.86, "text": "只有OPUS模型在这项测试中拿过10分", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 289.86, "end": 291.8, "text": "而像CLUD3.5这样的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 291.8, "end": 293.62, "text": "和KIMMYK3也只有9分", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 293.62, "end": 295.54, "text": "所以这开局只能算重归重矩", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 295.54, "end": 298.64, "text": "第二个测试是FreeJF隐形眼镜盒", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 298.64, "end": 301.24, "text": "它需要生成一个3D模型", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 301.24, "end": 304.94, "text": "是一个带有明显L和R标识盖子的隐形眼镜盒", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 304.94, "end": 308.66, "text": "并且点击盖子能够将其打开这项测试", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 308.66, "end": 311.28, "text": "几乎所有模型都会在这里翻车", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 311.28, "end": 312.48, "text": "我们发送过去看看", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 312.48, "end": 314.8, "text": "好的 这其实做得非常好", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 314.8, "end": 316.14, "text": "它拿到了8分", "chunk": 0, "language": "zh", "avg_logprob": -0.12761690261516165, "no_speech_prob": 1.1978875356921659e-11, "compression_ratio": 1.2507552870090635, "source": "whisper" }, { "start": 316.14, "end": 317.88, "text": "盒子看起来很像样", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 317.88, "end": 319.3, "text": "像个合格的眼镜盒L", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 319.3, "end": 321.8, "text": "而且L和二盖也清晰可见", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 321.8, "end": 323.24, "text": "点击打开的交互也有效", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 323.24, "end": 325.08, "text": "这让他与Quan2.5Maps并列", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 325.08, "end": 326.42, "text": "而Opus拿到了5分", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 326.42, "end": 327.34, "text": "是历史第二高分", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 327.34, "end": 329.52, "text": "在该测试中仅次于Plus 3.5", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 329.52, "end": 331.14, "text": "他仍然是唯一在这个测试中", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 331.14, "end": 332.18, "text": "拿到满分10分的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 332.18, "end": 333.86, "text": "所以对于一个小模型来说", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 333.86, "end": 335.64, "text": "能有这样的表现已经非常惊艳了", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 335.64, "end": 337.12, "text": "第三题是", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 337.12, "end": 338.58, "text": "FreeJF折叠桌", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 338.58, "end": 339.86, "text": "这里有一个滑块", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 339.86, "end": 342.04, "text": "当你向右拖动它时", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 342.04, "end": 343.4, "text": "桌子应该展开", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 343.4, "end": 344.78, "text": "向左拖则折叠回去", "chunk": 0, "language": "zh", "avg_logprob": -0.13947994337169403, "no_speech_prob": 1.6769415717154956e-11, "compression_ratio": 1.2248677248677249, "source": "whisper" }, { "start": 344.78, "end": 346.48, "text": "动画需要无缝衔剑", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 346.48, "end": 348.5, "text": "且呈现出真正的3D效果", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 348.5, "end": 349.78, "text": "我们发送请求看看", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 349.78, "end": 351.26, "text": "这个得到了7分", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 351.26, "end": 352.78, "text": "桌子确实能折叠起来", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 352.78, "end": 355.22, "text": "并能随滑快拖动而展开", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 355.22, "end": 356.82, "text": "过度基本自然", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 356.82, "end": 358.4, "text": "只是还不够完全流畅", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 358.4, "end": 360.16, "text": "表现最好的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 360.16, "end": 361.16, "text": "如Cloud 3.5", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 361.16, "end": 362.34, "text": "KIMI GLM4", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 362.34, "end": 363.8, "text": "以及Summit 3.5", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 363.8, "end": 365.44, "text": "在这项测试中都拿到了9分", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 365.44, "end": 366.64, "text": "但对比来看", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 366.64, "end": 368.46, "text": "Opus在这项测试中仅得了5分", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 368.46, "end": 370.74, "text": "所以这表现绝对算很不错了", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 370.74, "end": 373.06, "text": "第四题是生成熊猫的SVG", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 373.06, "end": 374.22, "text": "吃着汉堡", "chunk": 0, "language": "zh", "avg_logprob": -0.1372820910285501, "no_speech_prob": 1.4449250823611592e-11, "compression_ratio": 1.1807909604519775, "source": "whisper" }, { "start": 374.22, "end": 375.42, "text": "我们发送过去看看", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 375.42, "end": 376.5, "text": "好", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 376.5, "end": 378.62, "text": "这确实能看出是一只熊猫", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 378.62, "end": 380.42, "text": "也能看出是汉堡", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 380.42, "end": 381.86, "text": "但构图有点别扭", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 381.86, "end": 384.62, "text": "而且看起来不太像在吃汉堡", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 384.62, "end": 386.02, "text": "这题给五分", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 386.02, "end": 388.0, "text": "在这项测试中", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 388.0, "end": 391.6, "text": "像Pimi和Gemini 1.5这样表现最好的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 391.6, "end": 395.24, "text": "以及Quain 2.5 Max得分也仅为8分", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 395.24, "end": 397.34, "text": "所以并不算惊艳", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 397.34, "end": 399.18, "text": "第五题是制作弓箭", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 399.18, "end": 401.94, "text": "模拟游戏里面设计了四个靶子", "chunk": 0, "language": "zh", "avg_logprob": -0.13202353968764796, "no_speech_prob": 1.0998798226358186e-11, "compression_ratio": 1.137123745819398, "source": "whisper" }, { "start": 401.94, "end": 404.7, "text": "玩家需要以最短的时间击中所有靶子", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 404.7, "end": 406.12, "text": "从而登上排行榜", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 406.12, "end": 407.74, "text": "我们发送代码运行一下看看", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 407.74, "end": 409.14, "text": "这道题得了7分", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 409.14, "end": 410.96, "text": "核心射击机制可以正常运行", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 410.96, "end": 412.66, "text": "靶子能判定击中", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 412.66, "end": 414.04, "text": "排行榜功能也正常", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 414.04, "end": 416.2, "text": "但整体手感和细节打磨", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 416.2, "end": 417.84, "text": "还达不到顶尖模型的水平", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 417.84, "end": 420.74, "text": "比如Grog 1.5和Quang 2.5 Math", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 420.74, "end": 423.56, "text": "虽然Opus模型在这里都拿了满分", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 423.56, "end": 425.1, "text": "但这依然是个可玩的游戏", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 425.1, "end": 426.96, "text": "所以表现还算不错", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 426.96, "end": 429.24, "text": "第六题是一道数学难题", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 429.24, "end": 431.32, "text": "关于计算有序队的排列数", "chunk": 0, "language": "zh", "avg_logprob": -0.10077982041442278, "no_speech_prob": 1.3044664307071407e-11, "compression_ratio": 1.247282608695652, "source": "whisper" }, { "start": 431.32, "end": 433.3, "text": "答应该是2460", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 433.3, "end": 435.74, "text": "很多模型在这一题上完全翻车", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 435.74, "end": 436.62, "text": "我们提交一下", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 436.62, "end": 437.7, "text": "看看", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 437.7, "end": 439.86, "text": "他完美答对了2460", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 439.86, "end": 441.42, "text": "完全正确", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 441.42, "end": 442.64, "text": "拿了满分10分", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 442.64, "end": 444.3, "text": "这让他机身同一梯队", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 444.3, "end": 446.16, "text": "与Cloud 3.5", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 446.16, "end": 446.8, "text": "Timmy Chat", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 446.8, "end": 448.06, "text": "Queen Max", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 448.06, "end": 449.4, "text": "以及Opus模型并列", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 449.4, "end": 450.06, "text": "有趣的是", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 450.06, "end": 451.6, "text": "在我测试时", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 451.6, "end": 453.76, "text": "V4 Pro预览版也做对了这道题", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 453.76, "end": 457.32, "text": "所以V4系列的推理能力显然非常过瘾", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 457.32, "end": 460.6, "text": "第七题是一个复杂的长城任务", "chunk": 0, "language": "zh", "avg_logprob": -0.12382130450512989, "no_speech_prob": 1.1701846956702067e-11, "compression_ratio": 1.1253918495297806, "source": "whisper" }, { "start": 460.6, "end": 463.58, "text": "生成一个关于熊猫知识的数据集", "chunk": 0, "language": "zh", "avg_logprob": -0.10998057111909118, "no_speech_prob": 1.576908395528598e-11, "compression_ratio": 1.330935251798561, "source": "whisper" }, { "start": 463.58, "end": 466.88, "text": "并用该数据集为调一个GEM2B模型", "chunk": 0, "language": "zh", "avg_logprob": -0.10998057111909118, "no_speech_prob": 1.576908395528598e-11, "compression_ratio": 1.330935251798561, "source": "whisper" }, { "start": 466.88, "end": 471.24, "text": "接着构建一个本地WebUI每次刷新页面都会生成一条熊猫冷知识", "chunk": 0, "language": "zh", "avg_logprob": -0.10998057111909118, "no_speech_prob": 1.576908395528598e-11, "compression_ratio": 1.330935251798561, "source": "whisper" }, { "start": 471.24, "end": 473.5, "text": "全程从零开始完全自主完成", "chunk": 0, "language": "zh", "avg_logprob": -0.10998057111909118, "no_speech_prob": 1.576908395528598e-11, "compression_ratio": 1.330935251798561, "source": "whisper" }, { "start": 473.5, "end": 475.18, "text": "且完全在本地运行", "chunk": 0, "language": "zh", "avg_logprob": -0.10998057111909118, "no_speech_prob": 1.576908395528598e-11, "compression_ratio": 1.330935251798561, "source": "whisper" }, { "start": 475.18, "end": 476.5, "text": "我们启动任务看看效果", "chunk": 0, "language": "zh", "avg_logprob": -0.10998057111909118, "no_speech_prob": 1.576908395528598e-11, "compression_ratio": 1.330935251798561, "source": "whisper" }, { "start": 476.5, "end": 478.56, "text": "这次的表现绝对可以达满分", "chunk": 0, "language": "zh", "avg_logprob": -0.10998057111909118, "no_speech_prob": 1.576908395528598e-11, "compression_ratio": 1.330935251798561, "source": "whisper" }, { "start": 478.56, "end": 481.04, "text": "它生成了数据集并完成了整个微调", "chunk": 0, "language": "zh", "avg_logprob": -0.10998057111909118, "no_speech_prob": 1.576908395528598e-11, "compression_ratio": 1.330935251798561, "source": "whisper" }, { "start": 481.04, "end": 483.26, "text": "独自运行期间完全没有卡科", "chunk": 0, "language": "zh", "avg_logprob": -0.10998057111909118, "no_speech_prob": 1.576908395528598e-11, "compression_ratio": 1.330935251798561, "source": "whisper" }, { "start": 483.672, "end": 485.532, "text": "生成的Web UI正常运行", "chunk": 1, "language": "zh", "avg_logprob": -0.09404984745410604, "no_speech_prob": 2.0994801383511508e-11, "compression_ratio": 1.3491620111731844, "source": "whisper" }, { "start": 485.532, "end": 488.052, "text": "每次刷新都显示新的熊猫冷知识", "chunk": 1, "language": "zh", "avg_logprob": -0.09404984745410604, "no_speech_prob": 2.0994801383511508e-11, "compression_ratio": 1.3491620111731844, "source": "whisper" }, { "start": 488.052, "end": 491.372, "text": "这正是那种复杂的长程智能体任务", "chunk": 1, "language": "zh", "avg_logprob": -0.09404984745410604, "no_speech_prob": 2.0994801383511508e-11, "compression_ratio": 1.3491620111731844, "source": "whisper" }, { "start": 491.372, "end": 493.392, "text": "Deep C声称对其进行过针对性升级", "chunk": 1, "language": "zh", "avg_logprob": -0.09404984745410604, "no_speech_prob": 2.0994801383511508e-11, "compression_ratio": 1.3491620111731844, "source": "whisper" }, { "start": 493.392, "end": 494.472, "text": "说实话", "chunk": 1, "language": "zh", "avg_logprob": -0.09404984745410604, "no_speech_prob": 2.0994801383511508e-11, "compression_ratio": 1.3491620111731844, "source": "whisper" }, { "start": 494.472, "end": 495.432, "text": "效果显而易见", "chunk": 1, "language": "zh", "avg_logprob": -0.09404984745410604, "no_speech_prob": 2.0994801383511508e-11, "compression_ratio": 1.3491620111731844, "source": "whisper" }, { "start": 495.432, "end": 497.192, "text": "之前的预览版远没这么可靠", "chunk": 1, "language": "zh", "avg_logprob": -0.09404984745410604, "no_speech_prob": 2.0994801383511508e-11, "compression_ratio": 1.3491620111731844, "source": "whisper" }, { "start": 497.192, "end": 499.372, "text": "这波表现确实堪称完美", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 499.372, "end": 501.032, "text": "第八题也是最后一题", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 501.032, "end": 502.112, "text": "是制作一个3D万秒", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 502.112, "end": 503.992, "text": "要求是一个功能完整的3D万秒", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 503.992, "end": 505.392, "text": "能显示当前实际时间", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 505.392, "end": 506.312, "text": "秒针 分针", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 506.312, "end": 507.832, "text": "以及时针都要平滑转动", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 507.832, "end": 508.492, "text": "还要有日期", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 508.492, "end": 509.492, "text": "核心器显示", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 509.492, "end": 510.112, "text": "以及双时区", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 510.112, "end": 511.632, "text": "这是整个测试集中最难的一题", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 511.632, "end": 513.172, "text": "之前没有任何模型", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 513.172, "end": 514.572, "text": "能在这道题上拿到好成绩", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 514.572, "end": 516.712, "text": "历史最高分是Fibo5创下的", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 516.712, "end": 517.452, "text": "4分", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 517.452, "end": 518.612, "text": "让我们直接运行测试", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 518.612, "end": 519.672, "text": "但你感谢", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 519.672, "end": 521.572, "text": "Lip6 V4 Flav竟然拿到了6分", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 521.572, "end": 523.212, "text": "创下这道题历史记录", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 523.212, "end": 524.772, "text": "手表完整呈现出来", "chunk": 1, "language": "zh", "avg_logprob": -0.14845343695746527, "no_speech_prob": 3.143774476854766e-11, "compression_ratio": 1.3064516129032258, "source": "whisper" }, { "start": 527.192, "end": 528.272, "text": "指针转动平滑", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 528.272, "end": 529.552, "text": "只能实时同步时间", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 529.552, "end": 531.352, "text": "双时区设置也基本正常", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 531.352, "end": 532.172, "text": "虽然还不完美", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 532.172, "end": 534.152, "text": "所以没能拿到更高的分数", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 534.152, "end": 536.112, "text": "但这个小模型刚刚击败了", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 536.112, "end": 537.092, "text": "Cloud 3.5 Opus", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 537.092, "end": 539.272, "text": "以及我测试集最难题目上的", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 539.272, "end": 540.332, "text": "其他所有前沿模型", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 540.332, "end": 542.092, "text": "这完全出乎我的意料", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 542.092, "end": 543.892, "text": "来看一下最终的图表", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 543.892, "end": 545.932, "text": "Deep Seek V4 Flash 0731", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 545.932, "end": 547.912, "text": "最终拿到了58分", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 547.912, "end": 550.052, "text": "得分率为72.5%", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 550.052, "end": 550.712, "text": "这让它排在", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 550.712, "end": 555.272, "text": "略高于GPT4O513的71.25%", "chunk": 1, "language": "zh", "avg_logprob": -0.1564563835523405, "no_speech_prob": 1.709903052482531e-11, "compression_ratio": 1.1323529411764706, "source": "whisper" }, { "start": 555.272, "end": 558.852, "text": "稍低于Gemini 1.5 Pro的75%", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 558.852, "end": 559.752, "text": "而在它之上", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 559.752, "end": 561.172, "text": "有Kinning和", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 561.172, "end": 563.352, "text": "Opas并列77.5%", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 563.352, "end": 565.192, "text": "GTT 450808达80%", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 565.192, "end": 565.832, "text": "接着是Quain", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 565.832, "end": 569.152, "text": "Quain 2.5 Max拿下81.25%", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 569.152, "end": 572.032, "text": "而Cloud 3.5仍以82.5%稳居榜首", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 572.032, "end": 573.572, "text": "而对比以下背景", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 573.572, "end": 575.092, "text": "你就能明白这有多惊人", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 575.092, "end": 577.732, "text": "我之前测试Deep Seek V4 Pro预览版时", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 577.732, "end": 580.532, "text": "它在此测试中仅得了24.8%", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 580.532, "end": 584.092, "text": "新的Flash模型得分达到了72.5%", "chunk": 1, "language": "zh", "avg_logprob": -0.2157912587010583, "no_speech_prob": 1.4283847542262418e-11, "compression_ratio": 1.0620915032679739, "source": "whisper" }, { "start": 584.092, "end": 585.892, "text": "这意味着该系列的小模型现在", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 585.892, "end": 588.472, "text": "比大模型预览版的表现提升了近三倍", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 588.472, "end": 589.872, "text": "在我的测试中", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 589.872, "end": 592.572, "text": "而且在架构和尺寸与预览版相同", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 592.572, "end": 594.892, "text": "单靠后训练就能取得惊人提升", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 594.892, "end": 595.472, "text": "不可思议", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 595.472, "end": 599.052, "text": "这也印证了Deep Seed所宣称的Agent能力升级", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 599.052, "end": 601.932, "text": "从得分规律中也可以清楚地看到", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 601.932, "end": 603.712, "text": "所有射击推理", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 603.712, "end": 606.892, "text": "或长城Agent任务基本上都拿到了满分", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 606.892, "end": 609.292, "text": "比如数学和微调任务", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 609.292, "end": 611.532, "text": "还有创纪录的时钟生成常识", "chunk": 1, "language": "zh", "avg_logprob": -0.10888650569509953, "no_speech_prob": 1.6499329680286223e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 611.532, "end": 614.112, "text": "但像熊猫SVG这种纯视觉打磨的任务", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 614.112, "end": 615.252, "text": "还有电梯动画", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 615.252, "end": 616.892, "text": "在这些方面它的表现依然平平", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 616.892, "end": 618.592, "text": "所以这确实很像是一个", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 618.592, "end": 620.372, "text": "针对智能体编程深度微调的模型", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 620.372, "end": 622.292, "text": "正如官方基准测试所示", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 622.292, "end": 623.292, "text": "那么结论是什么", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 623.292, "end": 625.532, "text": "这绝对是目前性价比最高的模型之一", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 625.532, "end": 626.772, "text": "可用于智能体编程", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 626.772, "end": 627.792, "text": "在我的测试中", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 627.792, "end": 629.692, "text": "它的表现已经达到了Gemini 1.5 Pro的水平", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 629.692, "end": 632.012, "text": "它可以直接接入Cursor等代码工具", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 632.012, "end": 633.392, "text": "这得益于其API知识", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 633.392, "end": 634.972, "text": "而且Ditsig的价格", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 634.972, "end": 636.312, "text": "一直都非常便宜", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 636.312, "end": 637.112, "text": "如果你是学生", "chunk": 1, "language": "zh", "avg_logprob": -0.09499048868815105, "no_speech_prob": 2.1858773471827853e-11, "compression_ratio": 1.2417061611374407, "source": "whisper" }, { "start": 637.112, "end": 639.092, "text": "或者对于预算有限的开发者", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 639.092, "end": 640.412, "text": "这绝对是个极佳的选择", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 640.412, "end": 642.612, "text": "他显然还无法击败Cloud 3.5", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 642.612, "end": 644.572, "text": "或是在综合表现上超越OPUS", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 644.572, "end": 646.672, "text": "而且他的视觉能力依然表现平平", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 646.672, "end": 648.252, "text": "但考虑到他的体量和价格", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 648.252, "end": 649.292, "text": "这已经足够经验了", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 649.292, "end": 650.032, "text": "而且别忘了", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 650.032, "end": 651.672, "text": "这还仅仅是Flash版本", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 651.672, "end": 653.412, "text": "官方的DeepSick V4 Pro也即将发布", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 653.412, "end": 654.652, "text": "据官方介绍", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 654.652, "end": 656.492, "text": "如果将同样的智能体后训练应用", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 656.492, "end": 658.452, "text": "到1.6万亿参数的模型上", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 658.452, "end": 660.452, "text": "那对市面上昂贵的B元模型来说", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 660.452, "end": 661.812, "text": "将是降维打击般的存在", "chunk": 1, "language": "zh", "avg_logprob": -0.11794892205132379, "no_speech_prob": 2.0463349764687777e-11, "compression_ratio": 1.196642685851319, "source": "whisper" }, { "start": 661.812, "end": 663.932, "text": "我会在上线后第一时间进行测试", "chunk": 1, "language": "zh", "avg_logprob": -0.13653454909453522, "no_speech_prob": 5.066837574507765e-12, "compression_ratio": 0.8764044943820225, "source": "whisper" }, { "start": 663.932, "end": 664.592, "text": "尽情关注", "chunk": 1, "language": "zh", "avg_logprob": -0.13653454909453522, "no_speech_prob": 5.066837574507765e-12, "compression_ratio": 0.8764044943820225, "source": "whisper" }, { "start": 664.592, "end": 665.692, "text": "总的来说", "chunk": 1, "language": "zh", "avg_logprob": -0.13653454909453522, "no_speech_prob": 5.066837574507765e-12, "compression_ratio": 0.8764044943820225, "source": "whisper" }, { "start": 665.692, "end": 666.432, "text": "这非常酷", "chunk": 1, "language": "zh", "avg_logprob": -0.13653454909453522, "no_speech_prob": 5.066837574507765e-12, "compression_ratio": 0.8764044943820225, "source": "whisper" } ], "transcript_selection": { "schema": "videonote.transcript_selection.v1", "status": "SELECTED", "requested_language": null, "duration": 666.900313, "reference_path": null, "reference_kind": null, "reference_language": null, "selected_source": "whisper_segmented", "decision_reason": "source_unavailable_or_untrusted_whisper_passed", "whisper_rejected": false, "source_profile": { "segment_count": 0, "duration_coverage": 0.0, "usable_full": false, "usable_partial": false, "hallucination_region_count": 0, "hallucination_regions": [], "language": { "requested": null, "matched": false } }, "whisper_profile": { "segment_count": 343, "multilingual": false, "first_start": 0.0, "final_end": 666.432, "duration_coverage": 0.999298, "maximum_gap_seconds": 2.42, "inverted_count": 0, "language": { "requested": null, "matched": true, "latin": 644, "han": 2984, "kana": 0, "hangul": 0 }, "hallucination_region_count": 0, "hallucination_regions": [], "usable_full": true, "usable_partial": true }, "comparison": { "window_seconds": 30.0, "compared_window_count": 0, "median_score": 0.0, "divergent_window_count": 0, "divergent_ratio": 0.0, "divergent_windows": [] }, "timestamp_alignment": { "applied": false, "estimated_offset_seconds": 0.0, "matched_segment_count": 0, "match_scores": [] }, "filled_source_profile": null, "whisper_duration_guard": { "media_duration": 666.9, "rejected_segment_count": 0, "rejected_segments": [] } } }