{ "text": "OpenAI released GPT 6.1 SOL on September 29.\nSome say this model slashed costs by 50% for their $200 tier.\nOthers say it unceremoniously cut their subscription usage by 50% while calling it an efficiency boost.\nWhy are there two completely opposite reactions to the same announcement?\nBecause cheaper, and, more expensive, happened on the same day.\nDepending on whether you pay per token or per month.\nThis isn't about whether the new model is good.\nIt's about OpenAI saying two contradictory things to different users.\nToday, we won't discuss if 6.1 is worth using.\nWe'll clarify three things.\nFirst, why can the bills for models with the same list price differ by 10 times?\nSecond, which official data matches independent evaluations, and which do not?\nThird, how should you determine if it is right for you?\nOkay, let's start with the first one.\nGPT-6.1 is not the flagship model.\nIt is an upgraded version of the GPT-6 SOL released on September 22nd.\nOpenAI's current product line looks like this.\nAt the top, GPT-6 Ultra is the most powerful and expensive.\nIn the middle, GPT-6 SOL is slightly weaker and cheaper.\nAt the bottom, GPT-6 Lite is the weakest and cheapest.\nThe official positioning for 6.1 SOL is this.\nProviding Near Ultra Intelligence at 1-5th the price of Ultra.\nThe pricing is as follows.\nUltra is $10 input, $50 output\n6.1 SOL is $2 input, $10 output\nLight is $0.10 input, $0.50 output\nIf you calculate Ultra divided by SOL, it is exactly 5 times\nThis is what the official statement means by 1 fifth the price\nBut that is just the unit price\nThe real cost is a completely different story\nWhy?\nOn September 29th, someone on X posted a chart with a headline meaning that for the same $2\n$10.00 price tag, the bill can differ by 10 times. What were they comparing?\nGPT 6.1 SOL at $2 input, $10 output. Claude Sonnet 5.5 at $2 input, $10 output.\nThe prices are identical, but the per-task cost calculated by artificial analysis shows 5.5\nis $7.60, while 6.1 SOL is $0.72. A difference of 10.6 times. Why? Because Sonnet 5.5 is\n很 wordy\n它映出了一共193,000典的登場在本項上\n那是最高的一種的模式測試驗\n它是60%更多的 Opus 5.5\n和7倍的GPT6Ultra\n6.1 SOL\nA發現它的登場是10%-30%的高度\n但它是10%的一共1%的上一共1%\n它是比方的一页价值,\n但是因为它说少\n你费同一项供应价\n但一项供应价格更\n而另一项供应价格更多\n当然,这些人会比较不同\n这就是第一个我们必须理解\n在这些人的价格上不符定在这些价格上\n与其中的价格出现\n我们看看这些 data\n2从公司的公司\n2从公司的公司\n4个人数在 total\n第一\n75.2 vs 74.1\nThere is a chart on OpenAI's release page for SWE Bench.\nIt measures AI's ability to complete complex software engineering tasks in real codebases.\nO1 has five reasoning levels, low, medium, high, ultra, and max.\nThe numbers are, low is 70.0% at $0.17.\nMedium is 73.2% at $0.42.\nHigh is 75.2% at $0.65.\nUltra is 74.9% at $0.79\nMax is 71.9% at $1.57\nYou see, the high level is the best\nThe max level actually regressed\nAnd it costs 2.4 times more than the high level\nCompared to Claude 3.5 Sonnet's best at 74.1% for $4.43\nO1's high level scores higher and is 6.8 times cheaper\nThe official claim says it's, approaching, Sonnet\nbut the numbers show it surpasses Sonnet.\nThe official statement downplayed the result.\nThe second number, 71.9% equals 71.9%.\nThis is O1's ultra and max levels.\nThe scores are identical, both 71.9%, but max costs double.\nOfficial data tells you, don't set it to max.\nThis is also true on the Artificial Analysis Leaderboard.\ntheir coding agent index shows the ultra level is three points higher than max\nhigher reasoning intensity isn't always better\nthe third number 52 versus 58\nfrom artificial analysis's intelligence index\nthis is their comprehensive score from actual evaluations\nthe scores are\nclaude 3.5 opus at 58\nsonnet 3.5 at 56\nextra at 53\nflash at 53\nand 01 at 52.01 is the lowest.\nBut the cost is also the lowest.\nPer task, it's 72 cents.\nSonnet is $3.26, and Opus 3.5 is $5.98.\nClaude 3.5 Sonnet is $7.60.\nSo the question isn't which is best,\nbut whether your task needs a 58 or if a 52 is enough.\nThe fourth number, 646 vs 643 vs 728.\nThis is the big code bench leaderboard score\nNot an official list, but an independent one\nThe results are like this\n01-646 points\nPrevious version, 643 points\nGPT40, 728 points\n01 is only 3 points higher than the previous version\nAnd it lags behind 40 by 82 points\nThey outright say it moved a lot on their own benchmarks\nBut barely moved on independent ones\nThe term used is\nBenchmarking to the test on their own list\nCognition's Frontier Code 1.1 Extended is the same.\n01 is 60.4% and the previous version was 60.7%, almost flat.\nThe official and independent leaderboards tell two different stories.\nOK, pause.\nWe now know four numbers.\nFirst, it beats Sonnet on some tasks.\nSecond, Max is more expensive and worse.\nThird, the comprehensive score is the lowest of the bunch, but the cost is also the lowest.\n四、インデピンチマーク インプーブメント\nWhat do these four numbers tell you when put together?\n6.1 Sonnet is not a cheap version of Opus.\nIt is an efficiency model for specific tasks.\nIt is not an all-rounder.\nIt is specialized.\nSo the question arises.\nHow do you decide if it's right for you?\nWe divide scenarios into three categories.\nCode fixing, batch tasks, and repetitive work.\nIf you have to fix bugs daily or run massive repetitive knowledge tasks.\n6.1 sonnet is perfect for this\na developer conducted an experiment\nthey put 105 bugs into 2 code repositories\nopus scored 45 and cost 33 dollars\n6.1 sonnet scored 44 and cost 6 dollars and 56 cents\nopus 3.5 scored 41.7 and cost 58 dollars\n6.1 sonnet nearly matched opus at one fifth the cost\nhowever they noted n equals 1\nmeaning they only did this experiment once\nOn the Click, Benchmark for White Collar Knowledge Work\n6.1 Sonnet scored 80.0, costing 18 cents\nClaud 3.5 scored 81.4, costing 89\nFor these tasks, using 6.1 sonnet is more than enough.\nDon't use Macs.\nThe second category, Creative Visuals, Games, and Complex Reasoning.\n如果您做3D scenes, visual effects, or game design, tasks needing imagination,\n6.1 Sonnet may not be enough.\nBridgeBench asked it to render a sunset seascape.\nThe conclusion was it looked like a sunset in a different price bracket,\nand the sea was barely rendered.\nSomeone posted a tweet that got over 1,200 likes.\nOPUS 3.5 is much better than GPT 6.1.\nSomeone else made a comparison video.\nWith the same, high effort setting, the conclusion was that\nOPUS 3 won and 50 Jae-on in quality, bugs, and color schemes.\nSONNET is only better in token efficiency.\nOthers have also used logic questions to test it.\n6.1 SONNET set to max failed.\nOPUS 3.5 set to medium passed easily.\nFor these tasks, stick to CLOD or OPUS.\nThe third category, scientific research.\nThe company admitted this themselves.\nOn the Terminal Bench Science benchmark, 6.1 SONNET is 57.0, while OPUS is 68.1%.\nthe official word is that the most challenging scientific research tasks should use opus\nfor these tasks use opus and don't try to save a few bucks\nalright now you know 6.1 sonnet isn't a cheap opus\nit's an efficiency tool for specific tasks\nif your task aligns it's great if not it's a trap\nand there's one more thing you need to know\nOn September 29, the same day, OpenAI's TBO posted a tweet.\nThe quota for the new Pro 200 plan was reduced from 20x to 10x.\nOld users temporarily retained the 20x limit.\nOn the same day, they launched Pro 500.\n$500 per month for a 25x quota.\nAPI users saw it as getting cheaper.\nSubscription users saw it as getting more expensive.\nSo this isn't just a new model release.\nThis is OpenAI repricing for different user segments.\ndo you understand now why some people are cheering while others are cursing\nbecause they are looking at two different things\none last thing if you ask me if 6.15 is worth using\nmy answer is don't ask if it's worth it ask if it's suitable\nit is suitable for code repair and batch tasks\nuse the higher tier and turn on max\nfor creative and complex reasoning\ncurrent tests show it can't beat claude\nContinue using Claude or Opus\nAs for scientific research, the official guide already told you\nOne more thing about using Opus\nIf you are a subscriber, your quota has been cut\nYou need to recalculate if it's enough for you\nIf it's not enough, either upgrade to Pro500\nOr switch to API and pay by token\nThis isn't a question of which model is stronger\nThis is a question of OpenAI's repricing\nThat's all for today's content\nIf you liked it, remember to like, follow, and share\nThis is my motivation to keep creating. See you next time!", "language": "zh", "duration": 640.754688, "segments": [ { "start": 0.0, "end": 4.72, "text": "OpenAI released GPT 6.1 SOL on September 29.", "chunk": 0, "language": "zh", "avg_logprob": -0.18867753681383634, "no_speech_prob": 4.2470038597031134e-10, "compression_ratio": 1.5067114093959733 }, { "start": 4.72, "end": 9.46, "text": "Some say this model slashed costs by 50% for their $200 tier.", "chunk": 0, "language": "zh", "avg_logprob": -0.18867753681383634, "no_speech_prob": 4.2470038597031134e-10, "compression_ratio": 1.5067114093959733 }, { "start": 9.46, "end": 15.18, "text": "Others say it unceremoniously cut their subscription usage by 50% while calling it an efficiency boost.", "chunk": 0, "language": "zh", "avg_logprob": -0.18867753681383634, "no_speech_prob": 4.2470038597031134e-10, "compression_ratio": 1.5067114093959733 }, { "start": 15.18, "end": 18.68, "text": "Why are there two completely opposite reactions to the same announcement?", "chunk": 0, "language": "zh", "avg_logprob": -0.18867753681383634, "no_speech_prob": 4.2470038597031134e-10, "compression_ratio": 1.5067114093959733 }, { "start": 18.68, "end": 22.58, "text": "Because cheaper, and, more expensive, happened on the same day.", "chunk": 0, "language": "zh", "avg_logprob": -0.18867753681383634, "no_speech_prob": 4.2470038597031134e-10, "compression_ratio": 1.5067114093959733 }, { "start": 22.58, "end": 25.2, "text": "Depending on whether you pay per token or per month.", "chunk": 0, "language": "zh", "avg_logprob": -0.18867753681383634, "no_speech_prob": 4.2470038597031134e-10, "compression_ratio": 1.5067114093959733 }, { "start": 25.2, "end": 27.32, "text": "This isn't about whether the new model is good.", "chunk": 0, "language": "zh", "avg_logprob": -0.18867753681383634, "no_speech_prob": 4.2470038597031134e-10, "compression_ratio": 1.5067114093959733 }, { "start": 27.32, "end": 31.34, "text": "It's about OpenAI saying two contradictory things to different users.", "chunk": 0, "language": "zh", "avg_logprob": -0.09842588133731131, "no_speech_prob": 7.010395952980475e-13, "compression_ratio": 1.5354609929078014 }, { "start": 31.56, "end": 34.42, "text": "Today, we won't discuss if 6.1 is worth using.", "chunk": 0, "language": "zh", "avg_logprob": -0.09842588133731131, "no_speech_prob": 7.010395952980475e-13, "compression_ratio": 1.5354609929078014 }, { "start": 34.82, "end": 36.06, "text": "We'll clarify three things.", "chunk": 0, "language": "zh", "avg_logprob": -0.09842588133731131, "no_speech_prob": 7.010395952980475e-13, "compression_ratio": 1.5354609929078014 }, { "start": 36.24, "end": 40.34, "text": "First, why can the bills for models with the same list price differ by 10 times?", "chunk": 0, "language": "zh", "avg_logprob": -0.09842588133731131, "no_speech_prob": 7.010395952980475e-13, "compression_ratio": 1.5354609929078014 }, { "start": 40.58, "end": 45.06, "text": "Second, which official data matches independent evaluations, and which do not?", "chunk": 0, "language": "zh", "avg_logprob": -0.09842588133731131, "no_speech_prob": 7.010395952980475e-13, "compression_ratio": 1.5354609929078014 }, { "start": 45.24, "end": 48.02, "text": "Third, how should you determine if it is right for you?", "chunk": 0, "language": "zh", "avg_logprob": -0.09842588133731131, "no_speech_prob": 7.010395952980475e-13, "compression_ratio": 1.5354609929078014 }, { "start": 48.44, "end": 50.4, "text": "Okay, let's start with the first one.", "chunk": 0, "language": "zh", "avg_logprob": -0.09842588133731131, "no_speech_prob": 7.010395952980475e-13, "compression_ratio": 1.5354609929078014 }, { "start": 50.6, "end": 53.26, "text": "GPT-6.1 is not the flagship model.", "chunk": 0, "language": "zh", "avg_logprob": -0.09842588133731131, "no_speech_prob": 7.010395952980475e-13, "compression_ratio": 1.5354609929078014 }, { "start": 53.26, "end": 57.8, "text": "It is an upgraded version of the GPT-6 SOL released on September 22nd.", "chunk": 0, "language": "zh", "avg_logprob": -0.17607765197753905, "no_speech_prob": 9.03581354703975e-13, "compression_ratio": 1.5833333333333333 }, { "start": 57.92, "end": 60.22, "text": "OpenAI's current product line looks like this.", "chunk": 0, "language": "zh", "avg_logprob": -0.17607765197753905, "no_speech_prob": 9.03581354703975e-13, "compression_ratio": 1.5833333333333333 }, { "start": 60.38, "end": 63.98, "text": "At the top, GPT-6 Ultra is the most powerful and expensive.", "chunk": 0, "language": "zh", "avg_logprob": -0.17607765197753905, "no_speech_prob": 9.03581354703975e-13, "compression_ratio": 1.5833333333333333 }, { "start": 64.3, "end": 67.78, "text": "In the middle, GPT-6 SOL is slightly weaker and cheaper.", "chunk": 0, "language": "zh", "avg_logprob": -0.17607765197753905, "no_speech_prob": 9.03581354703975e-13, "compression_ratio": 1.5833333333333333 }, { "start": 68.14, "end": 71.58, "text": "At the bottom, GPT-6 Lite is the weakest and cheapest.", "chunk": 0, "language": "zh", "avg_logprob": -0.17607765197753905, "no_speech_prob": 9.03581354703975e-13, "compression_ratio": 1.5833333333333333 }, { "start": 72.22, "end": 75.24, "text": "The official positioning for 6.1 SOL is this.", "chunk": 0, "language": "zh", "avg_logprob": -0.17607765197753905, "no_speech_prob": 9.03581354703975e-13, "compression_ratio": 1.5833333333333333 }, { "start": 75.66, "end": 78.98, "text": "Providing Near Ultra Intelligence at 1-5th the price of Ultra.", "chunk": 0, "language": "zh", "avg_logprob": -0.17607765197753905, "no_speech_prob": 9.03581354703975e-13, "compression_ratio": 1.5833333333333333 }, { "start": 79.5, "end": 80.82, "text": "The pricing is as follows.", "chunk": 0, "language": "zh", "avg_logprob": -0.17607765197753905, "no_speech_prob": 9.03581354703975e-13, "compression_ratio": 1.5833333333333333 }, { "start": 80.82, "end": 84.2, "text": "Ultra is $10 input, $50 output", "chunk": 0, "language": "zh", "avg_logprob": -0.1497981866200765, "no_speech_prob": 9.17889896034918e-13, "compression_ratio": 1.5275590551181102 }, { "start": 84.2, "end": 88.04, "text": "6.1 SOL is $2 input, $10 output", "chunk": 0, "language": "zh", "avg_logprob": -0.1497981866200765, "no_speech_prob": 9.17889896034918e-13, "compression_ratio": 1.5275590551181102 }, { "start": 88.04, "end": 91.86, "text": "Light is $0.10 input, $0.50 output", "chunk": 0, "language": "zh", "avg_logprob": -0.1497981866200765, "no_speech_prob": 9.17889896034918e-13, "compression_ratio": 1.5275590551181102 }, { "start": 91.86, "end": 95.7, "text": "If you calculate Ultra divided by SOL, it is exactly 5 times", "chunk": 0, "language": "zh", "avg_logprob": -0.1497981866200765, "no_speech_prob": 9.17889896034918e-13, "compression_ratio": 1.5275590551181102 }, { "start": 95.7, "end": 98.74, "text": "This is what the official statement means by 1 fifth the price", "chunk": 0, "language": "zh", "avg_logprob": -0.1497981866200765, "no_speech_prob": 9.17889896034918e-13, "compression_ratio": 1.5275590551181102 }, { "start": 98.74, "end": 100.46, "text": "But that is just the unit price", "chunk": 0, "language": "zh", "avg_logprob": -0.1497981866200765, "no_speech_prob": 9.17889896034918e-13, "compression_ratio": 1.5275590551181102 }, { "start": 100.46, "end": 103.92, "text": "The real cost is a completely different story", "chunk": 0, "language": "zh", "avg_logprob": -0.1497981866200765, "no_speech_prob": 9.17889896034918e-13, "compression_ratio": 1.5275590551181102 }, { "start": 103.92, "end": 105.1, "text": "Why?", "chunk": 0, "language": "zh", "avg_logprob": -0.1497981866200765, "no_speech_prob": 9.17889896034918e-13, "compression_ratio": 1.5275590551181102 }, { "start": 105.1, "end": 110.38, "text": "On September 29th, someone on X posted a chart with a headline meaning that for the same $2", "chunk": 0, "language": "zh", "avg_logprob": -0.1497981866200765, "no_speech_prob": 9.17889896034918e-13, "compression_ratio": 1.5275590551181102 }, { "start": 110.38, "end": 115.28, "text": "$10.00 price tag, the bill can differ by 10 times. What were they comparing?", "chunk": 0, "language": "zh", "avg_logprob": -0.1370930049730384, "no_speech_prob": 6.794670078316689e-13, "compression_ratio": 1.491150442477876 }, { "start": 116.08, "end": 124.48, "text": "GPT 6.1 SOL at $2 input, $10 output. Claude Sonnet 5.5 at $2 input, $10 output.", "chunk": 0, "language": "zh", "avg_logprob": -0.1370930049730384, "no_speech_prob": 6.794670078316689e-13, "compression_ratio": 1.491150442477876 }, { "start": 124.66, "end": 130.62, "text": "The prices are identical, but the per-task cost calculated by artificial analysis shows 5.5", "chunk": 0, "language": "zh", "avg_logprob": -0.1370930049730384, "no_speech_prob": 6.794670078316689e-13, "compression_ratio": 1.491150442477876 }, { "start": 130.62, "end": 140.36, "text": "is $7.60, while 6.1 SOL is $0.72. A difference of 10.6 times. Why? Because Sonnet 5.5 is", "chunk": 0, "language": "zh", "avg_logprob": -0.1370930049730384, "no_speech_prob": 6.794670078316689e-13, "compression_ratio": 1.491150442477876 }, { "start": 140.36, "end": 141.12, "text": "很 wordy", "chunk": 0, "language": "zh", "avg_logprob": -0.16699374728732638, "no_speech_prob": 1.294094497360232e-12, "compression_ratio": 7.344262295081967 }, { "start": 141.12, "end": 145.46, "text": "它映出了一共193,000典的登場在本項上", "chunk": 0, "language": "zh", "avg_logprob": -0.8160187445746527, "no_speech_prob": 6.062570692862146e-13, "compression_ratio": 2.3106796116504853 }, { "start": 145.46, "end": 148.82, "text": "那是最高的一種的模式測試驗", "chunk": 0, "language": "zh", "avg_logprob": -0.8160187445746527, "no_speech_prob": 6.062570692862146e-13, "compression_ratio": 2.3106796116504853 }, { "start": 148.82, "end": 151.16, "text": "它是60%更多的 Opus 5.5", "chunk": 0, "language": "zh", "avg_logprob": -0.8160187445746527, "no_speech_prob": 6.062570692862146e-13, "compression_ratio": 2.3106796116504853 }, { "start": 151.16, "end": 153.66, "text": "和7倍的GPT6Ultra", "chunk": 0, "language": "zh", "avg_logprob": -0.8160187445746527, "no_speech_prob": 6.062570692862146e-13, "compression_ratio": 2.3106796116504853 }, { "start": 153.66, "end": 155.28, "text": "6.1 SOL", "chunk": 0, "language": "zh", "avg_logprob": -0.8160187445746527, "no_speech_prob": 6.062570692862146e-13, "compression_ratio": 2.3106796116504853 }, { "start": 155.28, "end": 161.22, "text": "A發現它的登場是10%-30%的高度", "chunk": 0, "language": "zh", "avg_logprob": -0.8160187445746527, "no_speech_prob": 6.062570692862146e-13, "compression_ratio": 2.3106796116504853 }, { "start": 161.22, "end": 165.32, "text": "但它是10%的一共1%的上一共1%", "chunk": 0, "language": "zh", "avg_logprob": -0.8160187445746527, "no_speech_prob": 6.062570692862146e-13, "compression_ratio": 2.3106796116504853 }, { "start": 165.32, "end": 167.32, "text": "它是比方的一页价值,", "chunk": 0, "language": "zh", "avg_logprob": -0.3199560546875, "no_speech_prob": 6.901918273016783e-13, "compression_ratio": 10.169230769230769 }, { "start": 167.32, "end": 168.72, "text": "但是因为它说少", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 168.72, "end": 170.4, "text": "你费同一项供应价", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 170.4, "end": 171.68, "text": "但一项供应价格更", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 171.68, "end": 173.04, "text": "而另一项供应价格更多", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 173.04, "end": 175.04, "text": "当然,这些人会比较不同", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 175.04, "end": 177.2, "text": "这就是第一个我们必须理解", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 177.2, "end": 179.56, "text": "在这些人的价格上不符定在这些价格上", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 180.44, "end": 182.16, "text": "与其中的价格出现", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 182.16, "end": 183.36, "text": "我们看看这些 data", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 184.0, "end": 185.52, "text": "2从公司的公司", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 185.52, "end": 187.56, "text": "2从公司的公司", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 187.56, "end": 189.04, "text": "4个人数在 total", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 189.76, "end": 190.48, "text": "第一", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 190.48, "end": 193.68, "text": "75.2 vs 74.1", "chunk": 0, "language": "zh", "avg_logprob": -1.0485459364853895, "no_speech_prob": 7.462482238054791e-13, "compression_ratio": 1.4549356223175967 }, { "start": 193.68, "end": 197.26, "text": "There is a chart on OpenAI's release page for SWE Bench.", "chunk": 0, "language": "zh", "avg_logprob": -0.22300320579892113, "no_speech_prob": 5.396288342389421e-13, "compression_ratio": 1.385321100917431 }, { "start": 197.98, "end": 203.04, "text": "It measures AI's ability to complete complex software engineering tasks in real codebases.", "chunk": 0, "language": "zh", "avg_logprob": -0.22300320579892113, "no_speech_prob": 5.396288342389421e-13, "compression_ratio": 1.385321100917431 }, { "start": 203.34, "end": 208.94, "text": "O1 has five reasoning levels, low, medium, high, ultra, and max.", "chunk": 0, "language": "zh", "avg_logprob": -0.22300320579892113, "no_speech_prob": 5.396288342389421e-13, "compression_ratio": 1.385321100917431 }, { "start": 209.72, "end": 213.48, "text": "The numbers are, low is 70.0% at $0.17.", "chunk": 0, "language": "zh", "avg_logprob": -0.22300320579892113, "no_speech_prob": 5.396288342389421e-13, "compression_ratio": 1.385321100917431 }, { "start": 215.1, "end": 218.38, "text": "Medium is 73.2% at $0.42.", "chunk": 0, "language": "zh", "avg_logprob": -0.22300320579892113, "no_speech_prob": 5.396288342389421e-13, "compression_ratio": 1.385321100917431 }, { "start": 219.28, "end": 222.46, "text": "High is 75.2% at $0.65.", "chunk": 0, "language": "zh", "avg_logprob": -0.22300320579892113, "no_speech_prob": 5.396288342389421e-13, "compression_ratio": 1.385321100917431 }, { "start": 222.46, "end": 226.52, "text": "Ultra is 74.9% at $0.79", "chunk": 0, "language": "zh", "avg_logprob": -0.12932540630472117, "no_speech_prob": 4.971313616840278e-13, "compression_ratio": 1.456221198156682 }, { "start": 226.52, "end": 230.98, "text": "Max is 71.9% at $1.57", "chunk": 0, "language": "zh", "avg_logprob": -0.12932540630472117, "no_speech_prob": 4.971313616840278e-13, "compression_ratio": 1.456221198156682 }, { "start": 230.98, "end": 233.74, "text": "You see, the high level is the best", "chunk": 0, "language": "zh", "avg_logprob": -0.12932540630472117, "no_speech_prob": 4.971313616840278e-13, "compression_ratio": 1.456221198156682 }, { "start": 233.74, "end": 235.68, "text": "The max level actually regressed", "chunk": 0, "language": "zh", "avg_logprob": -0.12932540630472117, "no_speech_prob": 4.971313616840278e-13, "compression_ratio": 1.456221198156682 }, { "start": 235.68, "end": 238.42, "text": "And it costs 2.4 times more than the high level", "chunk": 0, "language": "zh", "avg_logprob": -0.12932540630472117, "no_speech_prob": 4.971313616840278e-13, "compression_ratio": 1.456221198156682 }, { "start": 238.42, "end": 243.98, "text": "Compared to Claude 3.5 Sonnet's best at 74.1% for $4.43", "chunk": 0, "language": "zh", "avg_logprob": -0.12932540630472117, "no_speech_prob": 4.971313616840278e-13, "compression_ratio": 1.456221198156682 }, { "start": 243.98, "end": 248.62, "text": "O1's high level scores higher and is 6.8 times cheaper", "chunk": 0, "language": "zh", "avg_logprob": -0.12932540630472117, "no_speech_prob": 4.971313616840278e-13, "compression_ratio": 1.456221198156682 }, { "start": 248.62, "end": 252.04, "text": "The official claim says it's, approaching, Sonnet", "chunk": 0, "language": "zh", "avg_logprob": -0.12932540630472117, "no_speech_prob": 4.971313616840278e-13, "compression_ratio": 1.456221198156682 }, { "start": 252.46, "end": 254.5, "text": "but the numbers show it surpasses Sonnet.", "chunk": 0, "language": "zh", "avg_logprob": -0.2318473108036002, "no_speech_prob": 9.178900044551352e-13, "compression_ratio": 1.5258215962441315 }, { "start": 255.18, "end": 257.22, "text": "The official statement downplayed the result.", "chunk": 0, "language": "zh", "avg_logprob": -0.2318473108036002, "no_speech_prob": 9.178900044551352e-13, "compression_ratio": 1.5258215962441315 }, { "start": 257.42, "end": 261.8, "text": "The second number, 71.9% equals 71.9%.", "chunk": 0, "language": "zh", "avg_logprob": -0.2318473108036002, "no_speech_prob": 9.178900044551352e-13, "compression_ratio": 1.5258215962441315 }, { "start": 261.8, "end": 264.12, "text": "This is O1's ultra and max levels.", "chunk": 0, "language": "zh", "avg_logprob": -0.2318473108036002, "no_speech_prob": 9.178900044551352e-13, "compression_ratio": 1.5258215962441315 }, { "start": 264.84, "end": 270.26, "text": "The scores are identical, both 71.9%, but max costs double.", "chunk": 0, "language": "zh", "avg_logprob": -0.2318473108036002, "no_speech_prob": 9.178900044551352e-13, "compression_ratio": 1.5258215962441315 }, { "start": 271.68, "end": 274.44, "text": "Official data tells you, don't set it to max.", "chunk": 0, "language": "zh", "avg_logprob": -0.2318473108036002, "no_speech_prob": 9.178900044551352e-13, "compression_ratio": 1.5258215962441315 }, { "start": 275.22, "end": 278.18, "text": "This is also true on the Artificial Analysis Leaderboard.", "chunk": 0, "language": "zh", "avg_logprob": -0.2318473108036002, "no_speech_prob": 9.178900044551352e-13, "compression_ratio": 1.5258215962441315 }, { "start": 278.18, "end": 283.06, "text": "their coding agent index shows the ultra level is three points higher than max", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 283.06, "end": 286.3, "text": "higher reasoning intensity isn't always better", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 286.3, "end": 289.48, "text": "the third number 52 versus 58", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 289.48, "end": 292.72, "text": "from artificial analysis's intelligence index", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 292.72, "end": 295.82, "text": "this is their comprehensive score from actual evaluations", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 295.82, "end": 298.6, "text": "the scores are", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 298.6, "end": 301.08, "text": "claude 3.5 opus at 58", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 301.08, "end": 303.64, "text": "sonnet 3.5 at 56", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 303.64, "end": 305.12, "text": "extra at 53", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 305.12, "end": 306.7, "text": "flash at 53", "chunk": 0, "language": "zh", "avg_logprob": -0.21359725620435632, "no_speech_prob": 5.067969698416275e-13, "compression_ratio": 1.5741626794258374 }, { "start": 306.7, "end": 309.8, "text": "and 01 at 52.01 is the lowest.", "chunk": 0, "language": "zh", "avg_logprob": -0.1794333162560927, "no_speech_prob": 9.39685178627525e-13, "compression_ratio": 1.4517766497461928 }, { "start": 311.3, "end": 313.44, "text": "But the cost is also the lowest.", "chunk": 0, "language": "zh", "avg_logprob": -0.1794333162560927, "no_speech_prob": 9.39685178627525e-13, "compression_ratio": 1.4517766497461928 }, { "start": 313.66, "end": 315.54, "text": "Per task, it's 72 cents.", "chunk": 0, "language": "zh", "avg_logprob": -0.1794333162560927, "no_speech_prob": 9.39685178627525e-13, "compression_ratio": 1.4517766497461928 }, { "start": 315.8, "end": 320.08, "text": "Sonnet is $3.26, and Opus 3.5 is $5.98.", "chunk": 0, "language": "zh", "avg_logprob": -0.1794333162560927, "no_speech_prob": 9.39685178627525e-13, "compression_ratio": 1.4517766497461928 }, { "start": 320.5, "end": 323.08, "text": "Claude 3.5 Sonnet is $7.60.", "chunk": 0, "language": "zh", "avg_logprob": -0.1794333162560927, "no_speech_prob": 9.39685178627525e-13, "compression_ratio": 1.4517766497461928 }, { "start": 323.62, "end": 325.5, "text": "So the question isn't which is best,", "chunk": 0, "language": "zh", "avg_logprob": -0.1794333162560927, "no_speech_prob": 9.39685178627525e-13, "compression_ratio": 1.4517766497461928 }, { "start": 325.78, "end": 329.14, "text": "but whether your task needs a 58 or if a 52 is enough.", "chunk": 0, "language": "zh", "avg_logprob": -0.1794333162560927, "no_speech_prob": 9.39685178627525e-13, "compression_ratio": 1.4517766497461928 }, { "start": 329.5, "end": 334.76, "text": "The fourth number, 646 vs 643 vs 728.", "chunk": 0, "language": "zh", "avg_logprob": -0.1794333162560927, "no_speech_prob": 9.39685178627525e-13, "compression_ratio": 1.4517766497461928 }, { "start": 334.76, "end": 337.18, "text": "This is the big code bench leaderboard score", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 337.18, "end": 339.74, "text": "Not an official list, but an independent one", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 339.74, "end": 341.12, "text": "The results are like this", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 341.12, "end": 343.66, "text": "01-646 points", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 343.66, "end": 346.58, "text": "Previous version, 643 points", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 346.58, "end": 349.5, "text": "GPT40, 728 points", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 349.5, "end": 352.4, "text": "01 is only 3 points higher than the previous version", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 352.4, "end": 355.06, "text": "And it lags behind 40 by 82 points", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 355.06, "end": 358.26, "text": "They outright say it moved a lot on their own benchmarks", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 358.26, "end": 360.52, "text": "But barely moved on independent ones", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 360.52, "end": 362.0, "text": "The term used is", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 362.0, "end": 364.54, "text": "Benchmarking to the test on their own list", "chunk": 0, "language": "zh", "avg_logprob": -0.1499376537419167, "no_speech_prob": 1.4272002633528014e-12, "compression_ratio": 1.628 }, { "start": 364.76, "end": 368.08, "text": "Cognition's Frontier Code 1.1 Extended is the same.", "chunk": 0, "language": "zh", "avg_logprob": -0.1848585898416084, "no_speech_prob": 5.29193117278226e-13, "compression_ratio": 1.4844961240310077 }, { "start": 368.68, "end": 374.16, "text": "01 is 60.4% and the previous version was 60.7%, almost flat.", "chunk": 0, "language": "zh", "avg_logprob": -0.1848585898416084, "no_speech_prob": 5.29193117278226e-13, "compression_ratio": 1.4844961240310077 }, { "start": 374.66, "end": 378.04, "text": "The official and independent leaderboards tell two different stories.", "chunk": 0, "language": "zh", "avg_logprob": -0.1848585898416084, "no_speech_prob": 5.29193117278226e-13, "compression_ratio": 1.4844961240310077 }, { "start": 378.48, "end": 379.42, "text": "OK, pause.", "chunk": 0, "language": "zh", "avg_logprob": -0.1848585898416084, "no_speech_prob": 5.29193117278226e-13, "compression_ratio": 1.4844961240310077 }, { "start": 380.38, "end": 381.62, "text": "We now know four numbers.", "chunk": 0, "language": "zh", "avg_logprob": -0.1848585898416084, "no_speech_prob": 5.29193117278226e-13, "compression_ratio": 1.4844961240310077 }, { "start": 382.22, "end": 384.44, "text": "First, it beats Sonnet on some tasks.", "chunk": 0, "language": "zh", "avg_logprob": -0.1848585898416084, "no_speech_prob": 5.29193117278226e-13, "compression_ratio": 1.4844961240310077 }, { "start": 385.08, "end": 387.5, "text": "Second, Max is more expensive and worse.", "chunk": 0, "language": "zh", "avg_logprob": -0.1848585898416084, "no_speech_prob": 5.29193117278226e-13, "compression_ratio": 1.4844961240310077 }, { "start": 387.7, "end": 392.22, "text": "Third, the comprehensive score is the lowest of the bunch, but the cost is also the lowest.", "chunk": 0, "language": "zh", "avg_logprob": -0.1848585898416084, "no_speech_prob": 5.29193117278226e-13, "compression_ratio": 1.4844961240310077 }, { "start": 392.22, "end": 395.26, "text": "四、インデピンチマーク インプーブメント", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 395.26, "end": 397.66, "text": "What do these four numbers tell you when put together?", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 397.66, "end": 400.54, "text": "6.1 Sonnet is not a cheap version of Opus.", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 400.54, "end": 403.1, "text": "It is an efficiency model for specific tasks.", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 403.1, "end": 404.86, "text": "It is not an all-rounder.", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 404.86, "end": 406.54, "text": "It is specialized.", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 406.54, "end": 408.7, "text": "So the question arises.", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 408.7, "end": 411.42, "text": "How do you decide if it's right for you?", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 411.42, "end": 414.3, "text": "We divide scenarios into three categories.", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 414.3, "end": 417.58, "text": "Code fixing, batch tasks, and repetitive work.", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 417.58, "end": 421.5, "text": "If you have to fix bugs daily or run massive repetitive knowledge tasks.", "chunk": 0, "language": "zh", "avg_logprob": -0.2981535640076129, "no_speech_prob": 6.559978797150701e-13, "compression_ratio": 1.4615384615384615 }, { "start": 421.5, "end": 423.9, "text": "6.1 sonnet is perfect for this", "chunk": 0, "language": "zh", "avg_logprob": -0.1601067134312221, "no_speech_prob": 1.0080861926470286e-12, "compression_ratio": 1.7198067632850242 }, { "start": 423.9, "end": 426.08, "text": "a developer conducted an experiment", "chunk": 0, "language": "zh", "avg_logprob": -0.1601067134312221, "no_speech_prob": 1.0080861926470286e-12, "compression_ratio": 1.7198067632850242 }, { "start": 426.08, "end": 429.86, "text": "they put 105 bugs into 2 code repositories", "chunk": 0, "language": "zh", "avg_logprob": -0.1601067134312221, "no_speech_prob": 1.0080861926470286e-12, "compression_ratio": 1.7198067632850242 }, { "start": 429.86, "end": 432.92, "text": "opus scored 45 and cost 33 dollars", "chunk": 0, "language": "zh", "avg_logprob": -0.1601067134312221, "no_speech_prob": 1.0080861926470286e-12, "compression_ratio": 1.7198067632850242 }, { "start": 432.92, "end": 437.34, "text": "6.1 sonnet scored 44 and cost 6 dollars and 56 cents", "chunk": 0, "language": "zh", "avg_logprob": -0.1601067134312221, "no_speech_prob": 1.0080861926470286e-12, "compression_ratio": 1.7198067632850242 }, { "start": 437.34, "end": 442.08, "text": "opus 3.5 scored 41.7 and cost 58 dollars", "chunk": 0, "language": "zh", "avg_logprob": -0.1601067134312221, "no_speech_prob": 1.0080861926470286e-12, "compression_ratio": 1.7198067632850242 }, { "start": 442.08, "end": 446.04, "text": "6.1 sonnet nearly matched opus at one fifth the cost", "chunk": 0, "language": "zh", "avg_logprob": -0.1601067134312221, "no_speech_prob": 1.0080861926470286e-12, "compression_ratio": 1.7198067632850242 }, { "start": 446.04, "end": 448.1, "text": "however they noted n equals 1", "chunk": 0, "language": "zh", "avg_logprob": -0.1601067134312221, "no_speech_prob": 1.0080861926470286e-12, "compression_ratio": 1.7198067632850242 }, { "start": 448.1, "end": 450.34, "text": "meaning they only did this experiment once", "chunk": 0, "language": "zh", "avg_logprob": -0.1601067134312221, "no_speech_prob": 1.0080861926470286e-12, "compression_ratio": 1.7198067632850242 }, { "start": 450.34, "end": 453.22, "text": "On the Click, Benchmark for White Collar Knowledge Work", "chunk": 0, "language": "zh", "avg_logprob": -0.23024673461914064, "no_speech_prob": 6.873709500893055e-13, "compression_ratio": 1.1428571428571428 }, { "start": 453.22, "end": 457.02, "text": "6.1 Sonnet scored 80.0, costing 18 cents", "chunk": 0, "language": "zh", "avg_logprob": -0.23024673461914064, "no_speech_prob": 6.873709500893055e-13, "compression_ratio": 1.1428571428571428 }, { "start": 457.02, "end": 460.74, "text": "Claud 3.5 scored 81.4, costing 89", "chunk": 0, "language": "zh", "avg_logprob": -0.23024673461914064, "no_speech_prob": 6.873709500893055e-13, "compression_ratio": 1.1428571428571428 }, { "start": 461.195, "end": 464.955, "text": "For these tasks, using 6.1 sonnet is more than enough.", "chunk": 1, "language": "zh", "avg_logprob": -0.26097195943196616, "no_speech_prob": 4.3652012560180253e-10, "compression_ratio": 1.443089430894309 }, { "start": 464.955, "end": 466.155, "text": "Don't use Macs.", "chunk": 1, "language": "zh", "avg_logprob": -0.26097195943196616, "no_speech_prob": 4.3652012560180253e-10, "compression_ratio": 1.443089430894309 }, { "start": 466.155, "end": 471.115, "text": "The second category, Creative Visuals, Games, and Complex Reasoning.", "chunk": 1, "language": "zh", "avg_logprob": -0.26097195943196616, "no_speech_prob": 4.3652012560180253e-10, "compression_ratio": 1.443089430894309 }, { "start": 471.115, "end": 476.875, "text": "如果您做3D scenes, visual effects, or game design, tasks needing imagination,", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 477.115, "end": 479.195, "text": "6.1 Sonnet may not be enough.", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 479.515, "end": 482.035, "text": "BridgeBench asked it to render a sunset seascape.", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 482.075, "end": 484.955, "text": "The conclusion was it looked like a sunset in a different price bracket,", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 485.115, "end": 486.435, "text": "and the sea was barely rendered.", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 486.715, "end": 489.515, "text": "Someone posted a tweet that got over 1,200 likes.", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 489.595, "end": 492.835, "text": "OPUS 3.5 is much better than GPT 6.1.", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 492.915, "end": 494.595, "text": "Someone else made a comparison video.", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 494.755, "end": 497.355, "text": "With the same, high effort setting, the conclusion was that", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 497.355, "end": 500.795, "text": "OPUS 3 won and 50 Jae-on in quality, bugs, and color schemes.", "chunk": 1, "language": "zh", "avg_logprob": -0.3576325225830078, "no_speech_prob": 7.638728516910775e-13, "compression_ratio": 1.4738372093023255 }, { "start": 500.795, "end": 503.115, "text": "SONNET is only better in token efficiency.", "chunk": 1, "language": "zh", "avg_logprob": -0.24477001598903111, "no_speech_prob": 7.03799919819087e-13, "compression_ratio": 1.4615384615384615 }, { "start": 503.115, "end": 505.555, "text": "Others have also used logic questions to test it.", "chunk": 1, "language": "zh", "avg_logprob": -0.24477001598903111, "no_speech_prob": 7.03799919819087e-13, "compression_ratio": 1.4615384615384615 }, { "start": 505.555, "end": 508.315, "text": "6.1 SONNET set to max failed.", "chunk": 1, "language": "zh", "avg_logprob": -0.24477001598903111, "no_speech_prob": 7.03799919819087e-13, "compression_ratio": 1.4615384615384615 }, { "start": 508.315, "end": 511.675, "text": "OPUS 3.5 set to medium passed easily.", "chunk": 1, "language": "zh", "avg_logprob": -0.24477001598903111, "no_speech_prob": 7.03799919819087e-13, "compression_ratio": 1.4615384615384615 }, { "start": 511.675, "end": 514.675, "text": "For these tasks, stick to CLOD or OPUS.", "chunk": 1, "language": "zh", "avg_logprob": -0.24477001598903111, "no_speech_prob": 7.03799919819087e-13, "compression_ratio": 1.4615384615384615 }, { "start": 514.675, "end": 517.875, "text": "The third category, scientific research.", "chunk": 1, "language": "zh", "avg_logprob": -0.24477001598903111, "no_speech_prob": 7.03799919819087e-13, "compression_ratio": 1.4615384615384615 }, { "start": 517.875, "end": 519.595, "text": "The company admitted this themselves.", "chunk": 1, "language": "zh", "avg_logprob": -0.24477001598903111, "no_speech_prob": 7.03799919819087e-13, "compression_ratio": 1.4615384615384615 }, { "start": 519.595, "end": 526.115, "text": "On the Terminal Bench Science benchmark, 6.1 SONNET is 57.0, while OPUS is 68.1%.", "chunk": 1, "language": "zh", "avg_logprob": -0.24477001598903111, "no_speech_prob": 7.03799919819087e-13, "compression_ratio": 1.4615384615384615 }, { "start": 526.115, "end": 531.895, "text": "the official word is that the most challenging scientific research tasks should use opus", "chunk": 1, "language": "zh", "avg_logprob": -0.1006430957628333, "no_speech_prob": 7.759746621302521e-13, "compression_ratio": 1.6097560975609757 }, { "start": 531.895, "end": 535.815, "text": "for these tasks use opus and don't try to save a few bucks", "chunk": 1, "language": "zh", "avg_logprob": -0.1006430957628333, "no_speech_prob": 7.759746621302521e-13, "compression_ratio": 1.6097560975609757 }, { "start": 535.815, "end": 541.375, "text": "alright now you know 6.1 sonnet isn't a cheap opus", "chunk": 1, "language": "zh", "avg_logprob": -0.1006430957628333, "no_speech_prob": 7.759746621302521e-13, "compression_ratio": 1.6097560975609757 }, { "start": 541.375, "end": 544.095, "text": "it's an efficiency tool for specific tasks", "chunk": 1, "language": "zh", "avg_logprob": -0.1006430957628333, "no_speech_prob": 7.759746621302521e-13, "compression_ratio": 1.6097560975609757 }, { "start": 544.095, "end": 549.655, "text": "if your task aligns it's great if not it's a trap", "chunk": 1, "language": "zh", "avg_logprob": -0.1006430957628333, "no_speech_prob": 7.759746621302521e-13, "compression_ratio": 1.6097560975609757 }, { "start": 549.655, "end": 552.115, "text": "and there's one more thing you need to know", "chunk": 1, "language": "zh", "avg_logprob": -0.1006430957628333, "no_speech_prob": 7.759746621302521e-13, "compression_ratio": 1.6097560975609757 }, { "start": 552.115, "end": 556.495, "text": "On September 29, the same day, OpenAI's TBO posted a tweet.", "chunk": 1, "language": "zh", "avg_logprob": -0.16800384521484374, "no_speech_prob": 7.432733898846133e-13, "compression_ratio": 1.5617977528089888 }, { "start": 556.615, "end": 560.195, "text": "The quota for the new Pro 200 plan was reduced from 20x to 10x.", "chunk": 1, "language": "zh", "avg_logprob": -0.16800384521484374, "no_speech_prob": 7.432733898846133e-13, "compression_ratio": 1.5617977528089888 }, { "start": 560.355, "end": 562.735, "text": "Old users temporarily retained the 20x limit.", "chunk": 1, "language": "zh", "avg_logprob": -0.16800384521484374, "no_speech_prob": 7.432733898846133e-13, "compression_ratio": 1.5617977528089888 }, { "start": 562.895, "end": 565.095, "text": "On the same day, they launched Pro 500.", "chunk": 1, "language": "zh", "avg_logprob": -0.16800384521484374, "no_speech_prob": 7.432733898846133e-13, "compression_ratio": 1.5617977528089888 }, { "start": 565.355, "end": 568.095, "text": "$500 per month for a 25x quota.", "chunk": 1, "language": "zh", "avg_logprob": -0.16800384521484374, "no_speech_prob": 7.432733898846133e-13, "compression_ratio": 1.5617977528089888 }, { "start": 568.655, "end": 570.675, "text": "API users saw it as getting cheaper.", "chunk": 1, "language": "zh", "avg_logprob": -0.16800384521484374, "no_speech_prob": 7.432733898846133e-13, "compression_ratio": 1.5617977528089888 }, { "start": 571.155, "end": 573.555, "text": "Subscription users saw it as getting more expensive.", "chunk": 1, "language": "zh", "avg_logprob": -0.16800384521484374, "no_speech_prob": 7.432733898846133e-13, "compression_ratio": 1.5617977528089888 }, { "start": 574.115, "end": 576.115, "text": "So this isn't just a new model release.", "chunk": 1, "language": "zh", "avg_logprob": -0.16800384521484374, "no_speech_prob": 7.432733898846133e-13, "compression_ratio": 1.5617977528089888 }, { "start": 577.095, "end": 579.995, "text": "This is OpenAI repricing for different user segments.", "chunk": 1, "language": "zh", "avg_logprob": -0.16800384521484374, "no_speech_prob": 7.432733898846133e-13, "compression_ratio": 1.5617977528089888 }, { "start": 579.995, "end": 584.615, "text": "do you understand now why some people are cheering while others are cursing", "chunk": 1, "language": "zh", "avg_logprob": -0.1250907029255782, "no_speech_prob": 8.29253332668145e-13, "compression_ratio": 1.7092511013215859 }, { "start": 584.615, "end": 587.335, "text": "because they are looking at two different things", "chunk": 1, "language": "zh", "avg_logprob": -0.1250907029255782, "no_speech_prob": 8.29253332668145e-13, "compression_ratio": 1.7092511013215859 }, { "start": 587.335, "end": 593.055, "text": "one last thing if you ask me if 6.15 is worth using", "chunk": 1, "language": "zh", "avg_logprob": -0.1250907029255782, "no_speech_prob": 8.29253332668145e-13, "compression_ratio": 1.7092511013215859 }, { "start": 593.055, "end": 596.515, "text": "my answer is don't ask if it's worth it ask if it's suitable", "chunk": 1, "language": "zh", "avg_logprob": -0.1250907029255782, "no_speech_prob": 8.29253332668145e-13, "compression_ratio": 1.7092511013215859 }, { "start": 596.515, "end": 600.915, "text": "it is suitable for code repair and batch tasks", "chunk": 1, "language": "zh", "avg_logprob": -0.1250907029255782, "no_speech_prob": 8.29253332668145e-13, "compression_ratio": 1.7092511013215859 }, { "start": 600.915, "end": 603.215, "text": "use the higher tier and turn on max", "chunk": 1, "language": "zh", "avg_logprob": -0.1250907029255782, "no_speech_prob": 8.29253332668145e-13, "compression_ratio": 1.7092511013215859 }, { "start": 603.215, "end": 605.335, "text": "for creative and complex reasoning", "chunk": 1, "language": "zh", "avg_logprob": -0.1250907029255782, "no_speech_prob": 8.29253332668145e-13, "compression_ratio": 1.7092511013215859 }, { "start": 605.335, "end": 607.655, "text": "current tests show it can't beat claude", "chunk": 1, "language": "zh", "avg_logprob": -0.1250907029255782, "no_speech_prob": 8.29253332668145e-13, "compression_ratio": 1.7092511013215859 }, { "start": 607.655, "end": 609.695, "text": "Continue using Claude or Opus", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 609.695, "end": 612.955, "text": "As for scientific research, the official guide already told you", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 612.955, "end": 614.855, "text": "One more thing about using Opus", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 614.855, "end": 617.855, "text": "If you are a subscriber, your quota has been cut", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 617.855, "end": 620.255, "text": "You need to recalculate if it's enough for you", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 620.255, "end": 623.455, "text": "If it's not enough, either upgrade to Pro500", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 623.455, "end": 626.155, "text": "Or switch to API and pay by token", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 626.155, "end": 628.855, "text": "This isn't a question of which model is stronger", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 628.855, "end": 631.655, "text": "This is a question of OpenAI's repricing", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 631.655, "end": 633.455, "text": "That's all for today's content", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 633.455, "end": 636.655, "text": "If you liked it, remember to like, follow, and share", "chunk": 1, "language": "zh", "avg_logprob": -0.16827687863950377, "no_speech_prob": 7.639591541840074e-13, "compression_ratio": 1.583617747440273 }, { "start": 636.655, "end": 639.935, "text": "This is my motivation to keep creating. See you next time!", "chunk": 1, "language": "zh", "avg_logprob": -0.4662609398365021, "no_speech_prob": 4.684660320254663e-13, "compression_ratio": 0.9206349206349206 } ], "transcript_selection": { "schema": "videonote.transcript_selection.v1", "status": "SELECTED_WITH_WARNINGS", "selected_source": "whisper_provisional", "decision_reason": "published_with_warning_after_language_guard:requested_language_repair_inconclusive", "publication_allowed": true, "whisper_rejected": true, "publication_mode": "published_with_warnings" } }