{"text": "最近大模型竞赛又有新动静了,DeepSeek网页端被发现悄悄开启了新模型灰度测试,疑似正式传闻中的DeepSeek V4 Pro。从早期预览看,它在Web开发和复杂编程上表现很强,已经被拿来和顶级避原模型对照,但还得等正式发布后的独立测试和长期实测。对手也在优化体验,Anthropic官方给Cloud Code新增了Concise精简模式,直接给最终结果大幅削减冗余解释省Token。开源阵营同样在提速,Ornisk团队正式推出Ornisk 1.5系列开源大模型,从90亿密集型一直覆盖到3970亿的木屋架构。商业端的多项数据也在指向同一件事,企业订阅和AI编程工具仍在增长,模型能力正在从发布会话题变成团队每天都在用的生产力,真正关键不在热度而在使用频率。不止写代码,计算生物学也在往前走。Moderna与Merker的个性化癌症疫苗完成三期临床试验,至少说明AI正进入更高价值的研发协作环节。在看Deepseek的灰度迹象,开发者在响应里观察到了未公开模型的标志性短语,这是值得跟踪的信号,但还不能当成正式发布确认。在Web3D游戏和交互渲染测试里,它交出了很有竞争力的结果,真正值得看的是复杂前端生成正在成为模型比拼的新考场。生成、调试和迭代能否连成闭环才是下一关。SVG实量图形测试也给出了不错的观感,细节和质感被认为很突出。不过这类对比仍该看更多公开样例,而不是只看一张漂亮截图,数据和复现结果才更有说服力。而想要控制这套生成效率,只需在Cloud Code的配置文件里将Output Style参数设为Consise,就能自动砍掉多余废话,省下大笔Token。前面提到的开源模型,Onis 1.5也很聪明。它通过自己构建任务脚手架和自发生成测试题来进行强化学习,极大摆脱了对传统人工标注数据级的依赖。从公开基准成绩看,Onis 1.5已经拿到了很有分量的分数,开源模型离顶级商业能力还有多远,接下来要看更广泛,更可复现的评测。这类效率提升最终都会落到使用数据上。相关报告显示,AI变成工具在短时间内增长很快,开发者愿不愿意留下来,正在变成下一轮竞争的关键。资本市场的反应也很直接,Moderna公布相关三期结果后,股价明显波动。市场在压注的是AI能不能把新药研发从概念变成更可验证的效率,但股价不是技术价值的最终证明。国内API服务也在跟进。质朴开放GLM5.3的API,重点放在复杂agent,自动化编程和安全任务。对开发者来说,关键是把能力、价格和稳定性一起算。谷歌则在生态端大洒福利,面向全球高校学生推出返校技优惠。符合条件的大学生可以免费领一整年Google AI Pro以及Gemini进阶服务。我的判断是,前眼模型和生产力工具都得回到落地。感谢你看到这里,这里是大雷早上好,想一起练的。国内来星球,海外加入频道会员找大雷。", "segments": [{"id": 0, "seek": 0, "start": 0.0, "end": 10.28, "text": "最近大模型竞赛又有新动静了,DeepSeek网页端被发现悄悄开启了新模型灰度测试,疑似正式传闻中的DeepSeek V4 Pro。", "tokens": [50365, 46777, 3582, 41908, 39823, 11957, 252, 5266, 249, 17047, 2412, 12560, 34961, 5363, 247, 2289, 11, 11089, 595, 10637, 916, 16469, 239, 10178, 113, 11957, 107, 23238, 28926, 20204, 14696, 226, 14696, 226, 18937, 2392, 107, 2289, 12560, 41908, 39823, 13290, 108, 13127, 11038, 233, 5233, 243, 11, 18140, 239, 7384, 120, 15789, 27584, 7384, 254, 8259, 119, 5975, 1546, 11089, 595, 10637, 916, 691, 19, 1705, 1543, 50879], "temperature": 0, "avg_logprob": -0.05898970242204337, "compression_ratio": 1.1465201465201464, "no_speech_prob": 2.2121219786508384e-11}, {"id": 1, "seek": 0, "start": 10.44, "end": 21.580000000000002, "text": "从早期预览看,它在Web开发和复杂编程上表现很强,已经被拿来和顶级避原模型对照,但还得等正式发布后的独立测试和长期实测。", "tokens": [50887, 35630, 21176, 16786, 12501, 226, 6758, 230, 4200, 11, 11284, 3581, 4360, 65, 18937, 28926, 12565, 1787, 235, 4422, 224, 38109, 244, 29649, 5708, 17571, 20204, 4563, 5702, 118, 11, 49161, 23238, 24351, 6912, 12565, 10178, 114, 16853, 100, 3330, 123, 19683, 41908, 39823, 8713, 32150, 11, 8395, 14852, 5916, 10187, 15789, 27584, 28926, 34688, 13547, 1546, 18637, 105, 24409, 11038, 233, 5233, 243, 12565, 32271, 16786, 24726, 11038, 233, 1543, 51444], "temperature": 0, "avg_logprob": -0.05898970242204337, "compression_ratio": 1.1465201465201464, "no_speech_prob": 2.2121219786508384e-11}, {"id": 2, "seek": 2158, "start": 21.58, "end": 32.739999999999995, "text": "对手也在优化体验,Anthropic官方给Cloud Code新增了Concise精简模式,直接给最终结果大幅削减冗余解释省Token。", "tokens": [50365, 8713, 11389, 6404, 3581, 7384, 246, 23756, 29485, 49657, 234, 11, 7828, 14222, 299, 31929, 9249, 23197, 32787, 15549, 12560, 24228, 252, 2289, 9838, 66, 908, 34910, 11249, 222, 41908, 27584, 11, 43297, 23197, 8661, 10115, 230, 45641, 9319, 3582, 3509, 227, 5935, 232, 6336, 237, 5676, 245, 1593, 247, 17278, 5873, 232, 2862, 223, 51, 8406, 1543, 50923], "temperature": 0, "avg_logprob": -0.12198682469645823, "compression_ratio": 1.0449438202247192, "no_speech_prob": 1.1178444454806868e-11}, {"id": 3, "seek": 2158, "start": 32.9, "end": 43.94, "text": "开源阵营同样在提速,Ornisk团队正式推出Ornisk 1.5系列开源大模型,从90亿密集型一直覆盖到3970亿的木屋架构。", "tokens": [50931, 18937, 47402, 10034, 113, 10674, 98, 13089, 14496, 3581, 20949, 31217, 11, 21520, 77, 7797, 3919, 95, 10034, 253, 15789, 27584, 33597, 7781, 21520, 77, 7797, 502, 13, 20, 25368, 43338, 18937, 47402, 3582, 41908, 39823, 11, 35630, 7771, 1369, 123, 47838, 26020, 39823, 34448, 2888, 228, 5419, 244, 4511, 12493, 5867, 1369, 123, 1546, 41437, 9636, 233, 7360, 114, 7360, 226, 1543, 51483], "temperature": 0, "avg_logprob": -0.12198682469645823, "compression_ratio": 1.0449438202247192, "no_speech_prob": 1.1178444454806868e-11}, {"id": 4, "seek": 4394, "start": 43.94, "end": 59.32, "text": "商业端的多项数据也在指向同一件事,企业订阅和AI编程工具仍在增长,模型能力正在从发布会话题变成团队每天都在用的生产力,真正关键不在热度而在使用频率。", "tokens": [50365, 45581, 940, 248, 11957, 107, 1546, 6392, 10178, 117, 33188, 26075, 106, 6404, 3581, 25922, 24282, 13089, 2257, 20485, 6973, 11, 7384, 223, 940, 248, 7422, 95, 10034, 227, 12565, 48698, 38109, 244, 29649, 23323, 39806, 1550, 235, 3581, 24228, 252, 32271, 11, 41908, 39823, 8225, 13486, 15789, 3581, 35630, 28926, 34688, 12949, 21596, 30716, 2129, 246, 11336, 3919, 95, 10034, 253, 23664, 6135, 7182, 3581, 9254, 1546, 8244, 1369, 100, 13486, 11, 6303, 15789, 28053, 23049, 106, 1960, 3581, 23661, 255, 13127, 11070, 3581, 22982, 9254, 39752, 44866, 1543, 51134], "temperature": 0, "avg_logprob": -0.04526124998580578, "compression_ratio": 1.173780487804878, "no_speech_prob": 1.6396375576710476e-11}, {"id": 5, "seek": 4394, "start": 59.42, "end": 62.72, "text": "不止写代码,计算生物学也在往前走。", "tokens": [51139, 1960, 30438, 5676, 247, 19105, 23230, 223, 11, 7422, 94, 19497, 8244, 23516, 29618, 6404, 3581, 29510, 8945, 9575, 1543, 51304], "temperature": 0, "avg_logprob": -0.04526124998580578, "compression_ratio": 1.173780487804878, "no_speech_prob": 1.6396375576710476e-11}, {"id": 6, "seek": 4394, "start": 63.34, "end": 72.34, "text": "Moderna与Merker的个性化癌症疫苗完成三期临床试验,至少说明AI正进入更高价值的研发协作环节。", "tokens": [51335, 44, 19866, 629, 940, 236, 45757, 5767, 1546, 7549, 21686, 23756, 6651, 234, 15553, 229, 31670, 13736, 245, 41509, 10960, 16786, 940, 112, 6346, 232, 5233, 243, 49657, 234, 11, 20844, 15686, 8090, 11100, 48698, 15789, 36700, 14028, 19002, 12979, 1550, 115, 40242, 1546, 23230, 242, 28926, 5322, 237, 11914, 8051, 107, 45161, 1543, 51785], "temperature": 0, "avg_logprob": -0.04526124998580578, "compression_ratio": 1.173780487804878, "no_speech_prob": 1.6396375576710476e-11}, {"id": 7, "seek": 7234, "start": 72.34, "end": 83.60000000000001, "text": "在看Deepseek的灰度迹象,开发者在响应里观察到了未公开模型的标志性短语,这是值得跟踪的信号,但还不能当成正式发布确认。", "tokens": [50365, 3581, 4200, 11089, 595, 405, 916, 1546, 13290, 108, 13127, 3316, 117, 45007, 11, 18937, 28926, 12444, 3581, 3858, 235, 44297, 15759, 42350, 47550, 21381, 29954, 13545, 18937, 41908, 39823, 1546, 162, 3921, 44700, 21686, 5881, 255, 5233, 255, 11, 27455, 40242, 5916, 9678, 41166, 103, 1546, 17665, 26987, 11, 8395, 14852, 28590, 16233, 11336, 15789, 27584, 28926, 34688, 38114, 106, 7422, 97, 1543, 50928], "temperature": 0, "avg_logprob": -0.05456933268794307, "compression_ratio": 1.2193548387096773, "no_speech_prob": 1.6664905566621258e-11}, {"id": 8, "seek": 7234, "start": 83.60000000000001, "end": 94.62, "text": "在Web3D游戏和交互渲染测试里,它交出了很有竞争力的结果,真正值得看的是复杂前端生成正在成为模型比拼的新考场。", "tokens": [50928, 3581, 4360, 65, 18, 35, 9592, 116, 1486, 237, 12565, 28455, 1369, 240, 9592, 110, 46568, 11038, 233, 5233, 243, 15759, 11, 11284, 28455, 7781, 2289, 49295, 11957, 252, 1369, 231, 13486, 1546, 45641, 9319, 11, 6303, 15789, 40242, 5916, 4200, 24620, 1787, 235, 4422, 224, 8945, 11957, 107, 8244, 11336, 15789, 3581, 11336, 13992, 41908, 39823, 11706, 6852, 120, 1546, 12560, 26504, 50255, 1543, 51479], "temperature": 0, "avg_logprob": -0.05456933268794307, "compression_ratio": 1.2193548387096773, "no_speech_prob": 1.6664905566621258e-11}, {"id": 9, "seek": 7234, "start": 95.26, "end": 99.4, "text": "生成、调试和迭代能否连成闭环才是下一关。", "tokens": [51511, 8244, 11336, 1231, 8897, 225, 5233, 243, 12565, 3316, 255, 19105, 8225, 45334, 3316, 252, 11336, 8259, 255, 8051, 107, 18888, 1541, 45271, 28053, 1543, 51718], "temperature": 0, "avg_logprob": -0.05456933268794307, "compression_ratio": 1.2193548387096773, "no_speech_prob": 1.6664905566621258e-11}, {"id": 10, "seek": 9940, "start": 99.4, "end": 105.46000000000001, "text": "SVG实量图形测试也给出了不错的观感,细节和质感被认为很突出。", "tokens": [50365, 50, 53, 38, 24726, 26748, 3919, 122, 30900, 11038, 233, 5233, 243, 6404, 23197, 7781, 2289, 1960, 29900, 1546, 42350, 9709, 11, 10115, 228, 45161, 12565, 18464, 101, 9709, 23238, 7422, 97, 13992, 4563, 40859, 7781, 1543, 50668], "temperature": 0, "avg_logprob": -0.07468540748853363, "compression_ratio": 1.0837696335078535, "no_speech_prob": 1.88526382005616e-11}, {"id": 11, "seek": 9940, "start": 105.94000000000001, "end": 114.16000000000001, "text": "不过这类对比仍该看更多公开样例,而不是只看一张漂亮截图,数据和复现结果才更有说服力。", "tokens": [50692, 1960, 16866, 5562, 22113, 119, 8713, 11706, 1550, 235, 44646, 4200, 19002, 6392, 13545, 18937, 14496, 17797, 11, 11070, 7296, 14003, 4200, 2257, 44059, 29019, 1486, 103, 3919, 122, 11, 33188, 26075, 106, 12565, 1787, 235, 20204, 45641, 9319, 18888, 19002, 2412, 8090, 27408, 13486, 1543, 51103], "temperature": 0, "avg_logprob": -0.07468540748853363, "compression_ratio": 1.0837696335078535, "no_speech_prob": 1.88526382005616e-11}, {"id": 12, "seek": 11416, "start": 114.16, "end": 125.25999999999999, "text": "而想要控制这套生成效率,只需在Cloud Code的配置文件里将Output Style参数设为Consise,就能自动砍掉多余废话,省下大笔Token。", "tokens": [50365, 11070, 7093, 4275, 48707, 25491, 5562, 1881, 245, 8244, 11336, 43076, 44866, 11, 14003, 32535, 3581, 32787, 15549, 1546, 38846, 34719, 17174, 20485, 15759, 45456, 28353, 2582, 27004, 2129, 224, 33188, 7422, 122, 13992, 34, 892, 908, 11, 3111, 8225, 9722, 34961, 23230, 235, 29327, 6392, 1593, 247, 6346, 253, 21596, 11, 2862, 223, 4438, 3582, 5437, 242, 51, 8406, 1543, 50920], "temperature": 0, "avg_logprob": -0.07008807070843585, "compression_ratio": 1.084375, "no_speech_prob": 1.622444019411251e-11}, {"id": 13, "seek": 11416, "start": 125.38, "end": 128.38, "text": "前面提到的开源模型,Onis 1.5也很聪明。", "tokens": [50926, 8945, 8833, 20949, 4511, 1546, 18937, 47402, 41908, 39823, 11, 11747, 271, 502, 13, 20, 6404, 4563, 8171, 103, 11100, 1543, 51076], "temperature": 0, "avg_logprob": -0.07008807070843585, "compression_ratio": 1.084375, "no_speech_prob": 1.622444019411251e-11}, {"id": 14, "seek": 11416, "start": 128.8, "end": 138.35999999999999, "text": "它通过自己构建任务脚手架和自发生成测试题来进行强化学习,极大摆脱了对传统人工标注数据级的依赖。", "tokens": [51097, 11284, 19550, 16866, 17645, 7360, 226, 34157, 26443, 5087, 94, 27067, 248, 11389, 7360, 114, 12565, 9722, 28926, 8244, 11336, 11038, 233, 5233, 243, 30716, 6912, 36700, 8082, 5702, 118, 23756, 29618, 2930, 254, 11, 7360, 223, 3582, 34783, 228, 27067, 109, 2289, 8713, 7384, 254, 10115, 253, 4035, 23323, 162, 3921, 26432, 33188, 26075, 106, 16853, 100, 1546, 3254, 251, 5266, 244, 1543, 51575], "temperature": 0, "avg_logprob": -0.07008807070843585, "compression_ratio": 1.084375, "no_speech_prob": 1.622444019411251e-11}, {"id": 15, "seek": 13836, "start": 138.36, "end": 150.0, "text": "从公开基准成绩看,Onis 1.5已经拿到了很有分量的分数,开源模型离顶级商业能力还有多远,接下来要看更广泛,更可复现的评测。", "tokens": [50365, 35630, 13545, 18937, 26008, 6336, 228, 11336, 10115, 102, 4200, 11, 11747, 271, 502, 13, 20, 49161, 24351, 21381, 49295, 6627, 26748, 1546, 6627, 33188, 11, 18937, 47402, 41908, 39823, 18614, 119, 10178, 114, 16853, 100, 45581, 940, 248, 8225, 13486, 35091, 6392, 3316, 250, 11, 14468, 4438, 6912, 4275, 4200, 19002, 3509, 123, 6847, 249, 11, 19002, 4429, 1787, 235, 20204, 1546, 5233, 226, 11038, 233, 1543, 50947], "temperature": 0, "avg_logprob": -0.05179913334114842, "compression_ratio": 1.1924119241192412, "no_speech_prob": 1.449485323434807e-11}, {"id": 16, "seek": 13836, "start": 150.12, "end": 153.12, "text": "这类效率提升最终都会落到使用数据上。", "tokens": [50953, 5562, 22113, 119, 43076, 44866, 20949, 41670, 8661, 10115, 230, 7182, 12949, 30848, 4511, 22982, 9254, 33188, 26075, 106, 5708, 1543, 51103], "temperature": 0, "avg_logprob": -0.05179913334114842, "compression_ratio": 1.1924119241192412, "no_speech_prob": 1.449485323434807e-11}, {"id": 17, "seek": 13836, "start": 153.74, "end": 162.26000000000002, "text": "相关报告显示,AI变成工具在短时间内增长很快,开发者愿不愿意留下来,正在变成下一轮竞争的关键。", "tokens": [51134, 15106, 28053, 49817, 16846, 1431, 122, 25696, 11, 48698, 2129, 246, 11336, 23323, 39806, 3581, 5881, 255, 44848, 34742, 24228, 252, 32271, 4563, 10251, 11, 18937, 28926, 12444, 4820, 123, 1960, 4820, 123, 9042, 24456, 4438, 6912, 11, 15789, 3581, 2129, 246, 11336, 45271, 17819, 106, 11957, 252, 1369, 231, 1546, 28053, 23049, 106, 1543, 51560], "temperature": 0, "avg_logprob": -0.05179913334114842, "compression_ratio": 1.1924119241192412, "no_speech_prob": 1.449485323434807e-11}, {"id": 18, "seek": 13836, "start": 162.26000000000002, "end": 167.92000000000002, "text": "资本市场的反应也很直接,Moderna公布相关三期结果后,股价明显波动。", "tokens": [51560, 5266, 226, 8802, 27261, 50255, 1546, 22138, 44297, 6404, 4563, 43297, 11, 44, 19866, 629, 13545, 34688, 15106, 28053, 10960, 16786, 45641, 9319, 13547, 11, 14356, 94, 1550, 115, 11100, 1431, 122, 30806, 34961, 1543, 51843], "temperature": 0, "avg_logprob": -0.05179913334114842, "compression_ratio": 1.1924119241192412, "no_speech_prob": 1.449485323434807e-11}, {"id": 19, "seek": 16792, "start": 167.92, "end": 176.29999999999998, "text": "市场在压注的是AI能不能把新药研发从概念变成更可验证的效率,但股价不是技术价值的最终证明。", "tokens": [50365, 27261, 50255, 3581, 5014, 233, 26432, 24620, 48698, 8225, 28590, 16075, 12560, 31409, 107, 23230, 242, 28926, 35630, 26181, 33679, 2129, 246, 11336, 19002, 4429, 49657, 234, 5233, 223, 1546, 43076, 44866, 11, 8395, 14356, 94, 1550, 115, 7296, 32502, 1474, 107, 1550, 115, 40242, 1546, 8661, 10115, 230, 5233, 223, 11100, 1543, 50784], "temperature": 0, "avg_logprob": -0.08280607950773171, "compression_ratio": 1.1309090909090909, "no_speech_prob": 1.5802002067966114e-11}, {"id": 20, "seek": 16792, "start": 176.45999999999998, "end": 178.17999999999998, "text": "国内API服务也在跟进。", "tokens": [50792, 16086, 34742, 4715, 40, 27408, 5087, 94, 6404, 3581, 9678, 36700, 1543, 50878], "temperature": 0, "avg_logprob": -0.08280607950773171, "compression_ratio": 1.1309090909090909, "no_speech_prob": 1.5802002067966114e-11}, {"id": 21, "seek": 16792, "start": 179.01999999999998, "end": 185.54, "text": "质朴开放GLM5.3的API,重点放在复杂agent,自动化编程和安全任务。", "tokens": [50920, 18464, 101, 1474, 112, 18937, 12744, 19440, 44, 20, 13, 18, 1546, 4715, 40, 11, 12624, 12579, 12744, 3581, 1787, 235, 4422, 224, 559, 317, 11, 9722, 34961, 23756, 38109, 244, 29649, 12565, 45466, 26443, 5087, 94, 1543, 51246], "temperature": 0, "avg_logprob": -0.08280607950773171, "compression_ratio": 1.1309090909090909, "no_speech_prob": 1.5802002067966114e-11}, {"id": 22, "seek": 16792, "start": 186.0, "end": 190.72, "text": "对开发者来说,关键是把能力、价格和稳定性一起算。", "tokens": [51269, 8713, 18937, 28926, 12444, 6912, 8090, 11, 28053, 23049, 106, 1541, 16075, 8225, 13486, 1231, 1550, 115, 30921, 12565, 10415, 111, 12088, 21686, 29567, 19497, 1543, 51505], "temperature": 0, "avg_logprob": -0.08280607950773171, "compression_ratio": 1.1309090909090909, "no_speech_prob": 1.5802002067966114e-11}, {"id": 23, "seek": 19072, "start": 190.72, "end": 197.6, "text": "谷歌则在生态端大洒福利,面向全球高校学生推出返校技优惠。", "tokens": [50365, 8897, 115, 29582, 2437, 247, 3581, 8244, 3757, 223, 11957, 107, 3582, 12136, 240, 33952, 23700, 11, 8833, 24282, 11319, 28533, 12979, 49684, 29618, 8244, 33597, 7781, 3316, 242, 49684, 32502, 7384, 246, 4799, 254, 1543, 50709], "temperature": 0, "avg_logprob": -0.08035591670445033, "compression_ratio": 1.1226765799256506, "no_speech_prob": 1.6534705896131818e-11}, {"id": 24, "seek": 19072, "start": 198.14, "end": 204.7, "text": "符合条件的大学生可以免费领一整年Google AI Pro以及Gemini进阶服务。", "tokens": [50736, 5437, 99, 14245, 48837, 20485, 1546, 3582, 29618, 8244, 6723, 2347, 235, 18464, 117, 12501, 228, 2257, 27662, 5157, 12104, 3127, 7318, 1705, 40282, 38, 443, 3812, 36700, 10034, 114, 27408, 5087, 94, 1543, 51064], "temperature": 0, "avg_logprob": -0.08035591670445033, "compression_ratio": 1.1226765799256506, "no_speech_prob": 1.6534705896131818e-11}, {"id": 25, "seek": 19072, "start": 204.84, "end": 209.04, "text": "我的判断是,前眼模型和生产力工具都得回到落地。", "tokens": [51071, 14200, 2437, 97, 4307, 255, 1541, 11, 8945, 25281, 41908, 39823, 12565, 8244, 1369, 100, 13486, 23323, 39806, 7182, 5916, 8350, 4511, 30848, 10928, 1543, 51281], "temperature": 0, "avg_logprob": -0.08035591670445033, "compression_ratio": 1.1226765799256506, "no_speech_prob": 1.6534705896131818e-11}, {"id": 26, "seek": 19072, "start": 209.66, "end": 213.78, "text": "感谢你看到这里,这里是大雷早上好,想一起练的。", "tokens": [51312, 9709, 11340, 16529, 4511, 35102, 11, 35102, 1541, 3582, 6306, 115, 21176, 5708, 2131, 11, 7093, 29567, 10115, 225, 1546, 1543, 51518], "temperature": 0, "avg_logprob": -0.08035591670445033, "compression_ratio": 1.1226765799256506, "no_speech_prob": 1.6534705896131818e-11}, {"id": 27, "seek": 21378, "start": 213.78, "end": 216.98, "text": "国内来星球,海外加入频道会员找大雷。", "tokens": [50365, 16086, 34742, 6912, 20682, 28533, 11, 20078, 12022, 43878, 39752, 6025, 12949, 3606, 246, 25085, 3582, 6306, 115, 1543, 50525], "temperature": 0, "avg_logprob": -0.04418906298550693, "compression_ratio": 0.8253968253968254, "no_speech_prob": 1.559439036236121e-11}], "language": "zh"}