{
  "text": "So, looks like we just got flashed by\nGoogle DeepMind [music] once again\nbecause they just dropped Gemini 3.7\nFlash today. Yes, their newest model,\nwhich is once again a Flash model, is\nhere. And this is only after 3 weeks\nsince Gemini 3.6 Flash. And what they\nare saying is that Gemini 3.7 Flash is\ntheir most intelligent workhorse model.\nAnd this is coming 3 weeks after 3.6 six\nflash because of developer feedback and\nalgorithmic innovations. And that is why\nthey're saying the Gemini 3.7 Flash\nmodel is, you know, out with so much\nimprovement cuz if we take a look at the\nbenchmarks, this model is way better\nthan the Gemini 3.6 Flash. But I think\nthis is also because we know internally\nthat they're planning on cancelelling\nGemini 3.5 Pro completely and working on\nGemini 4 at the moment. So maybe the\nimprovements they made with the Pro\nmodel that they're not going to be\ndropping anymore. Maybe they repackaged\nit as a flash model because this model\nis better than 3.6 and if the timeline\nis three weeks plus they're planning on\ncancelling the 3.5 Pro, it's quite\nlikely that is the case. We'll just\nignore that and put that into the side.\nWe don't know when we're getting a new\nPro model from the Google Deep Mind\nteam, but this model across many of the\nbenchmarks we will see that yes, it is\nstronger than the 3.6 Flash model. For\nexample, when it comes to code quality,\nproduction code quality, the model is\nproducing 43.6%\nversus the flash model 34.4%.\nAnd then Sonet 5 is at 42.7%\nand the Terra model is at 41.3%.\nSo, one thing to remember, since this is\na flash model, you're not going to see\nthem compare this to Opus 5 or any of\nthe other stronger models. The reason\nbeing cuz this is not their strongest\ntier. and the 3.1 Pro model when they\nfinally decide to upgrade it to Gemini 4\nPro, then we'll see it being compared to\nOpus 5 or GPT 5.6 Soul. But anyways, we\ncan see that this model is an\nimprovement from 3.6 Flash and I would\nsay that's basically the biggest result\nor the biggest update that we saw with\nthe new model because it does not really\nlike change up things a lot is still\nlike not becoming the number one model\nor the number one flash model. Even on\nsome benchmarks, Deepseek version 4\nflash is actually cheaper and more\nintelligent than this model. But let's\njust take a look at the benchmarks that\nthey have published. On Long Horizon\nsoftware engineering, this model is a\nlittle bit behind GPT 5.6 Terra, which\nsits at 69.6%\nand then 65.3% is the Flash model. The\nolder Flash model 3.6 sits at 48.6%.\nSonic 5 sits at 53.8% 8%. And the new\nplayer that is finally being included on\nbenchmarks, which even the Google\nDeepMind team is considering with their\nlaunch, is Muse Spark 1.2, which is\nsitting at 54.9%.\nSo, welcome Meta to the benchmark charts\nbecause now we're starting to see it\nappear on more and more benchmarks. And\nif we also take a look at web\ndevelopment, this model is getting a ELO\nscore of 1588 versus their old model is\n1538. So, not a crazy difference, but\ncompared to everything else out there,\nthis is number one. When I say\neverything else out there, once again,\ncompared to all the mid-tier models,\nthis is beating all of them when it\ncomes to web development. Now, this\nmodel is probably going to be used in\nenterprises a lot just because of\nGoogle's footprint in the enterprise\nspace, but we are seeing this model\nachieve on the automation bench 30.4%.\nAnd GPT 5.6 Terra sits at 23.6%.\nSo, yes, this model is stronger than the\nother Flash models out there, but as I\nsaid, they haven't included Deepseek\nversion for Flash because if they do, in\nsome areas, that model is actually quite\nbetter than the 3.7 Flash model. Before\nwe continue, if you're building AI\nagents or just messing around with them,\nArcade is worth knowing about. It's the\nruntime that lets your agent actually do\nthings instead of just talking about\nthem because that's the gap right now.\nThe models are smart enough. Your agent\ncan figure out exactly what needs to\nhappen in your email, your Slack, your\nCRM. It just can't go in and do it. And\nthe reason isn't intelligence, it's\npermissions. Something has to prove the\nagent is allowed to act on behalf of a\nspecific person in a specific account.\nThat's the messy part everyone runs\ninto, and it's the partit actually\nhandles for you. So instead of your\nagent using one shared login for\neverybody, it acts as whoever is\nactually signed in with exactly the\naccess that person has. If they can't\nsee something, the agent can't either.\nAnd you never have to touch any of that\nsetup yourself. Then there's the tools.\nArcade has thousands of them already\nbuilt for Gmail, Google Drive, Slack,\nNotion, Salesforce, most of the apps\npeople already work in, and they are\nbuilt specifically for AI to use. So the\nagent gets it right the first time\ninstead of guessing and failing and\nretrying. It also keeps a record of\neverything, what the agent did, for who\nand where, which matters a lot the\nmoment other people start using the\nthing you built. And it works with\nwhatever you're already using. Any\nmodel, any framework, cloud, cursor,\nchat, GPT, doesn't matter. So you're not\njust giving an AI a list of tools and\n[music] hoping it works. You're giving\nit a place where it can safely take real\nactions in real apps. It's free to start\nand the link is in the description.\nThank you once again for Arcade for\nsponsoring today's video. Now, let's get\nback into the video. Now, one thing to\nnote is that the Gemini 3.7 Flash model\nthrough the end of this year, so end of\n2026, they have a cheap pricing model\nthat they're placing on the model. 75\nper 1 million input tokens and $3.75 for\n1 million output tokens. So, it's a\ncompetitive price, but this is only for\nthe next 6 months. Because after those\nsix months are done, the model's pricing\nis actually, you know, a little bit more\nexpensive. And now they show it at the\nbottom over here, you can see that after\nstarting January of 2027, it will become\n$1.50 per input and $7.50 per output. So\nyeah, it's still cheap compared to the\nother frontier labs, but it's not as\ncheap as, for example, MU Spark when the\npricing is updated or even the Deepsee\nversion for Flash. But across these\nbenchmarks, we can see that this model\nis better in many areas that they\nhighlighted at the top. But then they\nalso have some other areas like long\nvideo understanding, which this model\nexcels at 85.4% versus 78.9%\nfor the Terra model. And then the old\nmodel was also pretty good at that,\n84.2%.\n84.2%.\nThen long context performance the model\nis at 97% and this model the GPT 6 Terra\none it sits at 93.5%.\nSo yeah this model in summary it is\nbetter so it's not all negative but it's\nnot all like you know that positive\nwhere you are super excited for Google\nDeep Mind because as I said they're\nprobably still holding off their biggest\nrelease for Gemini 4 lineup. Now, one\nthing people might have missed in their\ncharts because these charts sometimes\nare so messy to read and understand, but\nthe 3.7 flash is worse than GPT 5.6\nLuna, which is the model over here,\nwhich is achieving a higher score on\nthis benchmark deepware engineering for\nabout three times the cost. So, yeah,\nthis is kind of interesting because\nyeah, the cost for Gemini 3.7 Flash is a\nlittle bit more than what it looks like.\nNow, some people are a little bit upset\nand they're like, \"Oh, disgraceful.\nGoogle left out soul, opus, and fable\nbecause Google is incredibly behind. But\nI think one thing we got to remember,\nguys, is that this model is a flash\nmodel. It's not trying to be a pro model\nor it's not trying to compete with Opus\nor Soul or Fable 5 level models yet.\nThat is probably going to be Gemini 4.\nSo when Gemini 4 comes out, then I think\nit's okay for us to criticize them if\nthey don't include Soul, Opus, and Fable\nin their benchmark charts because for\nnow, I think what they have done is\npretty accurate. One lab that I would\nhave liked to see or one model for\nexample I would have liked to see on\nthat chart would be Deepseek version 4\nFlash because that would kind of spoil\ntheir release because that model is way\ncheaper compared to the Gemini 3.7 flash\nmodel lineup. Now this model jumped from\nnumber 19 to 8 on the web development\narea and we see it over here now and\ncouple of models that are ahead of it\nare Opus 5 obviously Kim K3 Quinn 3.8\nMax Cloud Opus 5 Gro 4.6 6, which is a\nmodel that came yesterday, which was a\nbig win for SpaceX, Fable 5, and 5.6.\nSoul. So, yeah, this model is trying to\ncompete in the web development, but it's\nstill behind all of these models, which\nis, you know, expected cuz it's a flash\nmodel. It's not really a pro model.\nToday, OpenAI has also launched a weight\nlist for 5.6 so ultra fast mode. Now,\nthis is possible because of the fast\nchips that they have access to now\nthrough Cabus and GPT 5.6 6o with this\nchip kind of running it is able to\nachieve an ultra fast mode that\ngenerates up to 750 output tokens per\nsecond which is about 14 times faster\nthan the standard mode. So we're getting\na really fast version of GPT 5.6 so now\nthis is supposed to be used for live or\nnear production workloads like you know\nreal-time voice support commerce what\nthis allows GPT 5.6 six soul to do is\nkind of be really fast in critical\nsituations when people might be\ninteracting with the AI agent like\nfinancial research security response\nsupport I think is going to be a big\narea where this new ultraast mode will\nbe kind of implemented business and\ndeveloper agents maybe yeah but I think\nlike support or near production\nworkloads like real-time voice I see\nthis model really excelling at that and\nobviously this is still a weightless\nmode we don't know how many people are\ngoing to get access to this, how\nsuccessful it is or what the pricing is.\nI don't know if the pricing has changed\nbecause there's no information in their\nactual, you know, blog post. But as I\nmentioned, couple of areas where they\nmentioned that this is going to be\nreally important. Customer support and\nvoice, commerce, live research and\nexperimentation, financial research and\nsecurity, incident response and\nreliability. But yeah, I think this\npartnership is going to be important for\nOpenAI going forward. But as I said, 14\ntimes the speed. What does that mean for\ncost? We don't know yet. But that's it\nfor today's video. Make sure you guys\nare subscribed to the channel. Follow\nour new newsletter as well at\nuniverseai.behive.com\nuniverseai.behive.com\nas well as subscribe to the main channel\nWorld of AI and support us on X by\nfollowing the Universe of AIZ as well.\nUntil then, I'll see you guys in the\nnext",
  "language": "en",
  "duration": 633.16175,
  "segments": [
    {
      "start": 0.24,
      "end": 2.48,
      "text": "So, looks like we just got flashed by",
      "source": "source_caption"
    },
    {
      "start": 2.48,
      "end": 4.64,
      "text": "Google DeepMind [music] once again",
      "source": "source_caption"
    },
    {
      "start": 4.64,
      "end": 6.879,
      "text": "because they just dropped Gemini 3.7",
      "source": "source_caption"
    },
    {
      "start": 6.879,
      "end": 9.44,
      "text": "Flash today. Yes, their newest model,",
      "source": "source_caption"
    },
    {
      "start": 9.44,
      "end": 11.679,
      "text": "which is once again a Flash model, is",
      "source": "source_caption"
    },
    {
      "start": 11.679,
      "end": 14.16,
      "text": "here. And this is only after 3 weeks",
      "source": "source_caption"
    },
    {
      "start": 14.16,
      "end": 17.279,
      "text": "since Gemini 3.6 Flash. And what they",
      "source": "source_caption"
    },
    {
      "start": 17.279,
      "end": 20.0,
      "text": "are saying is that Gemini 3.7 Flash is",
      "source": "source_caption"
    },
    {
      "start": 20.0,
      "end": 22.8,
      "text": "their most intelligent workhorse model.",
      "source": "source_caption"
    },
    {
      "start": 22.8,
      "end": 25.359,
      "text": "And this is coming 3 weeks after 3.6 six",
      "source": "source_caption"
    },
    {
      "start": 25.359,
      "end": 27.84,
      "text": "flash because of developer feedback and",
      "source": "source_caption"
    },
    {
      "start": 27.84,
      "end": 30.48,
      "text": "algorithmic innovations. And that is why",
      "source": "source_caption"
    },
    {
      "start": 30.48,
      "end": 32.48,
      "text": "they're saying the Gemini 3.7 Flash",
      "source": "source_caption"
    },
    {
      "start": 32.48,
      "end": 34.32,
      "text": "model is, you know, out with so much",
      "source": "source_caption"
    },
    {
      "start": 34.32,
      "end": 36.16,
      "text": "improvement cuz if we take a look at the",
      "source": "source_caption"
    },
    {
      "start": 36.16,
      "end": 38.399,
      "text": "benchmarks, this model is way better",
      "source": "source_caption"
    },
    {
      "start": 38.399,
      "end": 41.36,
      "text": "than the Gemini 3.6 Flash. But I think",
      "source": "source_caption"
    },
    {
      "start": 41.36,
      "end": 43.44,
      "text": "this is also because we know internally",
      "source": "source_caption"
    },
    {
      "start": 43.44,
      "end": 44.879,
      "text": "that they're planning on cancelelling",
      "source": "source_caption"
    },
    {
      "start": 44.879,
      "end": 48.16,
      "text": "Gemini 3.5 Pro completely and working on",
      "source": "source_caption"
    },
    {
      "start": 48.16,
      "end": 50.48,
      "text": "Gemini 4 at the moment. So maybe the",
      "source": "source_caption"
    },
    {
      "start": 50.48,
      "end": 51.92,
      "text": "improvements they made with the Pro",
      "source": "source_caption"
    },
    {
      "start": 51.92,
      "end": 53.52,
      "text": "model that they're not going to be",
      "source": "source_caption"
    },
    {
      "start": 53.52,
      "end": 55.84,
      "text": "dropping anymore. Maybe they repackaged",
      "source": "source_caption"
    },
    {
      "start": 55.84,
      "end": 58.32,
      "text": "it as a flash model because this model",
      "source": "source_caption"
    },
    {
      "start": 58.32,
      "end": 61.199,
      "text": "is better than 3.6 and if the timeline",
      "source": "source_caption"
    },
    {
      "start": 61.199,
      "end": 63.12,
      "text": "is three weeks plus they're planning on",
      "source": "source_caption"
    },
    {
      "start": 63.12,
      "end": 65.519,
      "text": "cancelling the 3.5 Pro, it's quite",
      "source": "source_caption"
    },
    {
      "start": 65.519,
      "end": 67.6,
      "text": "likely that is the case. We'll just",
      "source": "source_caption"
    },
    {
      "start": 67.6,
      "end": 69.119,
      "text": "ignore that and put that into the side.",
      "source": "source_caption"
    },
    {
      "start": 69.119,
      "end": 70.479,
      "text": "We don't know when we're getting a new",
      "source": "source_caption"
    },
    {
      "start": 70.479,
      "end": 72.08,
      "text": "Pro model from the Google Deep Mind",
      "source": "source_caption"
    },
    {
      "start": 72.08,
      "end": 74.4,
      "text": "team, but this model across many of the",
      "source": "source_caption"
    },
    {
      "start": 74.4,
      "end": 77.119,
      "text": "benchmarks we will see that yes, it is",
      "source": "source_caption"
    },
    {
      "start": 77.119,
      "end": 79.92,
      "text": "stronger than the 3.6 Flash model. For",
      "source": "source_caption"
    },
    {
      "start": 79.92,
      "end": 81.84,
      "text": "example, when it comes to code quality,",
      "source": "source_caption"
    },
    {
      "start": 81.84,
      "end": 84.159,
      "text": "production code quality, the model is",
      "source": "source_caption"
    },
    {
      "start": 84.159,
      "end": 86.159,
      "text": "producing 43.6%",
      "source": "source_caption"
    },
    {
      "start": 86.159,
      "end": 89.68,
      "text": "versus the flash model 34.4%.",
      "source": "source_caption"
    },
    {
      "start": 89.68,
      "end": 93.52,
      "text": "And then Sonet 5 is at 42.7%",
      "source": "source_caption"
    },
    {
      "start": 93.52,
      "end": 96.799,
      "text": "and the Terra model is at 41.3%.",
      "source": "source_caption"
    },
    {
      "start": 96.799,
      "end": 98.799,
      "text": "So, one thing to remember, since this is",
      "source": "source_caption"
    },
    {
      "start": 98.799,
      "end": 100.24,
      "text": "a flash model, you're not going to see",
      "source": "source_caption"
    },
    {
      "start": 100.24,
      "end": 102.159,
      "text": "them compare this to Opus 5 or any of",
      "source": "source_caption"
    },
    {
      "start": 102.159,
      "end": 104.24,
      "text": "the other stronger models. The reason",
      "source": "source_caption"
    },
    {
      "start": 104.24,
      "end": 106.24,
      "text": "being cuz this is not their strongest",
      "source": "source_caption"
    },
    {
      "start": 106.24,
      "end": 109.119,
      "text": "tier. and the 3.1 Pro model when they",
      "source": "source_caption"
    },
    {
      "start": 109.119,
      "end": 111.6,
      "text": "finally decide to upgrade it to Gemini 4",
      "source": "source_caption"
    },
    {
      "start": 111.6,
      "end": 114.479,
      "text": "Pro, then we'll see it being compared to",
      "source": "source_caption"
    },
    {
      "start": 114.479,
      "end": 117.92,
      "text": "Opus 5 or GPT 5.6 Soul. But anyways, we",
      "source": "source_caption"
    },
    {
      "start": 117.92,
      "end": 119.28,
      "text": "can see that this model is an",
      "source": "source_caption"
    },
    {
      "start": 119.28,
      "end": 121.52,
      "text": "improvement from 3.6 Flash and I would",
      "source": "source_caption"
    },
    {
      "start": 121.52,
      "end": 124.56,
      "text": "say that's basically the biggest result",
      "source": "source_caption"
    },
    {
      "start": 124.56,
      "end": 126.479,
      "text": "or the biggest update that we saw with",
      "source": "source_caption"
    },
    {
      "start": 126.479,
      "end": 128.56,
      "text": "the new model because it does not really",
      "source": "source_caption"
    },
    {
      "start": 128.56,
      "end": 130.72,
      "text": "like change up things a lot is still",
      "source": "source_caption"
    },
    {
      "start": 130.72,
      "end": 132.56,
      "text": "like not becoming the number one model",
      "source": "source_caption"
    },
    {
      "start": 132.56,
      "end": 135.04,
      "text": "or the number one flash model. Even on",
      "source": "source_caption"
    },
    {
      "start": 135.04,
      "end": 137.2,
      "text": "some benchmarks, Deepseek version 4",
      "source": "source_caption"
    },
    {
      "start": 137.2,
      "end": 139.92,
      "text": "flash is actually cheaper and more",
      "source": "source_caption"
    },
    {
      "start": 139.92,
      "end": 142.16,
      "text": "intelligent than this model. But let's",
      "source": "source_caption"
    },
    {
      "start": 142.16,
      "end": 143.68,
      "text": "just take a look at the benchmarks that",
      "source": "source_caption"
    },
    {
      "start": 143.68,
      "end": 145.84,
      "text": "they have published. On Long Horizon",
      "source": "source_caption"
    },
    {
      "start": 145.84,
      "end": 148.08,
      "text": "software engineering, this model is a",
      "source": "source_caption"
    },
    {
      "start": 148.08,
      "end": 150.959,
      "text": "little bit behind GPT 5.6 Terra, which",
      "source": "source_caption"
    },
    {
      "start": 150.959,
      "end": 152.959,
      "text": "sits at 69.6%",
      "source": "source_caption"
    },
    {
      "start": 152.959,
      "end": 156.4,
      "text": "and then 65.3% is the Flash model. The",
      "source": "source_caption"
    },
    {
      "start": 156.4,
      "end": 160.8,
      "text": "older Flash model 3.6 sits at 48.6%.",
      "source": "source_caption"
    },
    {
      "start": 160.8,
      "end": 164.0,
      "text": "Sonic 5 sits at 53.8% 8%. And the new",
      "source": "source_caption"
    },
    {
      "start": 164.0,
      "end": 166.16,
      "text": "player that is finally being included on",
      "source": "source_caption"
    },
    {
      "start": 166.16,
      "end": 168.0,
      "text": "benchmarks, which even the Google",
      "source": "source_caption"
    },
    {
      "start": 168.0,
      "end": 169.92,
      "text": "DeepMind team is considering with their",
      "source": "source_caption"
    },
    {
      "start": 169.92,
      "end": 172.48,
      "text": "launch, is Muse Spark 1.2, which is",
      "source": "source_caption"
    },
    {
      "start": 172.48,
      "end": 174.64,
      "text": "sitting at 54.9%.",
      "source": "source_caption"
    },
    {
      "start": 174.64,
      "end": 177.519,
      "text": "So, welcome Meta to the benchmark charts",
      "source": "source_caption"
    },
    {
      "start": 177.519,
      "end": 179.36,
      "text": "because now we're starting to see it",
      "source": "source_caption"
    },
    {
      "start": 179.36,
      "end": 181.519,
      "text": "appear on more and more benchmarks. And",
      "source": "source_caption"
    },
    {
      "start": 181.519,
      "end": 182.8,
      "text": "if we also take a look at web",
      "source": "source_caption"
    },
    {
      "start": 182.8,
      "end": 185.36,
      "text": "development, this model is getting a ELO",
      "source": "source_caption"
    },
    {
      "start": 185.36,
      "end": 188.159,
      "text": "score of 1588 versus their old model is",
      "source": "source_caption"
    },
    {
      "start": 188.159,
      "end": 190.319,
      "text": "1538. So, not a crazy difference, but",
      "source": "source_caption"
    },
    {
      "start": 190.319,
      "end": 191.84,
      "text": "compared to everything else out there,",
      "source": "source_caption"
    },
    {
      "start": 191.84,
      "end": 193.36,
      "text": "this is number one. When I say",
      "source": "source_caption"
    },
    {
      "start": 193.36,
      "end": 194.879,
      "text": "everything else out there, once again,",
      "source": "source_caption"
    },
    {
      "start": 194.879,
      "end": 196.72,
      "text": "compared to all the mid-tier models,",
      "source": "source_caption"
    },
    {
      "start": 196.72,
      "end": 198.64,
      "text": "this is beating all of them when it",
      "source": "source_caption"
    },
    {
      "start": 198.64,
      "end": 200.64,
      "text": "comes to web development. Now, this",
      "source": "source_caption"
    },
    {
      "start": 200.64,
      "end": 202.48,
      "text": "model is probably going to be used in",
      "source": "source_caption"
    },
    {
      "start": 202.48,
      "end": 204.159,
      "text": "enterprises a lot just because of",
      "source": "source_caption"
    },
    {
      "start": 204.159,
      "end": 206.0,
      "text": "Google's footprint in the enterprise",
      "source": "source_caption"
    },
    {
      "start": 206.0,
      "end": 208.159,
      "text": "space, but we are seeing this model",
      "source": "source_caption"
    },
    {
      "start": 208.159,
      "end": 211.599,
      "text": "achieve on the automation bench 30.4%.",
      "source": "source_caption"
    },
    {
      "start": 211.599,
      "end": 215.44,
      "text": "And GPT 5.6 Terra sits at 23.6%.",
      "source": "source_caption"
    },
    {
      "start": 215.44,
      "end": 217.599,
      "text": "So, yes, this model is stronger than the",
      "source": "source_caption"
    },
    {
      "start": 217.599,
      "end": 219.76,
      "text": "other Flash models out there, but as I",
      "source": "source_caption"
    },
    {
      "start": 219.76,
      "end": 221.44,
      "text": "said, they haven't included Deepseek",
      "source": "source_caption"
    },
    {
      "start": 221.44,
      "end": 223.599,
      "text": "version for Flash because if they do, in",
      "source": "source_caption"
    },
    {
      "start": 223.599,
      "end": 225.519,
      "text": "some areas, that model is actually quite",
      "source": "source_caption"
    },
    {
      "start": 225.519,
      "end": 228.48,
      "text": "better than the 3.7 Flash model. Before",
      "source": "source_caption"
    },
    {
      "start": 228.48,
      "end": 230.319,
      "text": "we continue, if you're building AI",
      "source": "source_caption"
    },
    {
      "start": 230.319,
      "end": 232.48,
      "text": "agents or just messing around with them,",
      "source": "source_caption"
    },
    {
      "start": 232.48,
      "end": 234.799,
      "text": "Arcade is worth knowing about. It's the",
      "source": "source_caption"
    },
    {
      "start": 234.799,
      "end": 236.799,
      "text": "runtime that lets your agent actually do",
      "source": "source_caption"
    },
    {
      "start": 236.799,
      "end": 238.319,
      "text": "things instead of just talking about",
      "source": "source_caption"
    },
    {
      "start": 238.319,
      "end": 240.56,
      "text": "them because that's the gap right now.",
      "source": "source_caption"
    },
    {
      "start": 240.56,
      "end": 242.799,
      "text": "The models are smart enough. Your agent",
      "source": "source_caption"
    },
    {
      "start": 242.799,
      "end": 244.72,
      "text": "can figure out exactly what needs to",
      "source": "source_caption"
    },
    {
      "start": 244.72,
      "end": 246.56,
      "text": "happen in your email, your Slack, your",
      "source": "source_caption"
    },
    {
      "start": 246.56,
      "end": 249.68,
      "text": "CRM. It just can't go in and do it. And",
      "source": "source_caption"
    },
    {
      "start": 249.68,
      "end": 251.519,
      "text": "the reason isn't intelligence, it's",
      "source": "source_caption"
    },
    {
      "start": 251.519,
      "end": 253.68,
      "text": "permissions. Something has to prove the",
      "source": "source_caption"
    },
    {
      "start": 253.68,
      "end": 255.84,
      "text": "agent is allowed to act on behalf of a",
      "source": "source_caption"
    },
    {
      "start": 255.84,
      "end": 258.479,
      "text": "specific person in a specific account.",
      "source": "source_caption"
    },
    {
      "start": 258.479,
      "end": 260.239,
      "text": "That's the messy part everyone runs",
      "source": "source_caption"
    },
    {
      "start": 260.239,
      "end": 262.479,
      "text": "into, and it's the partit actually",
      "source": "source_caption"
    },
    {
      "start": 262.479,
      "end": 264.32,
      "text": "handles for you. So instead of your",
      "source": "source_caption"
    },
    {
      "start": 264.32,
      "end": 266.24,
      "text": "agent using one shared login for",
      "source": "source_caption"
    },
    {
      "start": 266.24,
      "end": 268.16,
      "text": "everybody, it acts as whoever is",
      "source": "source_caption"
    },
    {
      "start": 268.16,
      "end": 270.32,
      "text": "actually signed in with exactly the",
      "source": "source_caption"
    },
    {
      "start": 270.32,
      "end": 272.56,
      "text": "access that person has. If they can't",
      "source": "source_caption"
    },
    {
      "start": 272.56,
      "end": 274.88,
      "text": "see something, the agent can't either.",
      "source": "source_caption"
    },
    {
      "start": 274.88,
      "end": 276.479,
      "text": "And you never have to touch any of that",
      "source": "source_caption"
    },
    {
      "start": 276.479,
      "end": 278.96,
      "text": "setup yourself. Then there's the tools.",
      "source": "source_caption"
    },
    {
      "start": 278.96,
      "end": 280.8,
      "text": "Arcade has thousands of them already",
      "source": "source_caption"
    },
    {
      "start": 280.8,
      "end": 283.12,
      "text": "built for Gmail, Google Drive, Slack,",
      "source": "source_caption"
    },
    {
      "start": 283.12,
      "end": 285.52,
      "text": "Notion, Salesforce, most of the apps",
      "source": "source_caption"
    },
    {
      "start": 285.52,
      "end": 287.36,
      "text": "people already work in, and they are",
      "source": "source_caption"
    },
    {
      "start": 287.36,
      "end": 289.759,
      "text": "built specifically for AI to use. So the",
      "source": "source_caption"
    },
    {
      "start": 289.759,
      "end": 291.44,
      "text": "agent gets it right the first time",
      "source": "source_caption"
    },
    {
      "start": 291.44,
      "end": 293.04,
      "text": "instead of guessing and failing and",
      "source": "source_caption"
    },
    {
      "start": 293.04,
      "end": 295.12,
      "text": "retrying. It also keeps a record of",
      "source": "source_caption"
    },
    {
      "start": 295.12,
      "end": 297.52,
      "text": "everything, what the agent did, for who",
      "source": "source_caption"
    },
    {
      "start": 297.52,
      "end": 299.44,
      "text": "and where, which matters a lot the",
      "source": "source_caption"
    },
    {
      "start": 299.44,
      "end": 300.88,
      "text": "moment other people start using the",
      "source": "source_caption"
    },
    {
      "start": 300.88,
      "end": 302.639,
      "text": "thing you built. And it works with",
      "source": "source_caption"
    },
    {
      "start": 302.639,
      "end": 304.56,
      "text": "whatever you're already using. Any",
      "source": "source_caption"
    },
    {
      "start": 304.56,
      "end": 306.96,
      "text": "model, any framework, cloud, cursor,",
      "source": "source_caption"
    },
    {
      "start": 306.96,
      "end": 309.52,
      "text": "chat, GPT, doesn't matter. So you're not",
      "source": "source_caption"
    },
    {
      "start": 309.52,
      "end": 311.139,
      "text": "just giving an AI a list of tools and",
      "source": "source_caption"
    },
    {
      "start": 311.139,
      "end": 312.72,
      "text": "[music] hoping it works. You're giving",
      "source": "source_caption"
    },
    {
      "start": 312.72,
      "end": 314.8,
      "text": "it a place where it can safely take real",
      "source": "source_caption"
    },
    {
      "start": 314.8,
      "end": 317.36,
      "text": "actions in real apps. It's free to start",
      "source": "source_caption"
    },
    {
      "start": 317.36,
      "end": 319.12,
      "text": "and the link is in the description.",
      "source": "source_caption"
    },
    {
      "start": 319.12,
      "end": 320.88,
      "text": "Thank you once again for Arcade for",
      "source": "source_caption"
    },
    {
      "start": 320.88,
      "end": 322.96,
      "text": "sponsoring today's video. Now, let's get",
      "source": "source_caption"
    },
    {
      "start": 322.96,
      "end": 325.44,
      "text": "back into the video. Now, one thing to",
      "source": "source_caption"
    },
    {
      "start": 325.44,
      "end": 328.56,
      "text": "note is that the Gemini 3.7 Flash model",
      "source": "source_caption"
    },
    {
      "start": 328.56,
      "end": 330.32,
      "text": "through the end of this year, so end of",
      "source": "source_caption"
    },
    {
      "start": 330.32,
      "end": 333.36,
      "text": "2026, they have a cheap pricing model",
      "source": "source_caption"
    },
    {
      "start": 333.36,
      "end": 336.16,
      "text": "that they're placing on the model. 75",
      "source": "source_caption"
    },
    {
      "start": 336.16,
      "end": 339.84,
      "text": "per 1 million input tokens and $3.75 for",
      "source": "source_caption"
    },
    {
      "start": 339.84,
      "end": 341.68,
      "text": "1 million output tokens. So, it's a",
      "source": "source_caption"
    },
    {
      "start": 341.68,
      "end": 343.759,
      "text": "competitive price, but this is only for",
      "source": "source_caption"
    },
    {
      "start": 343.759,
      "end": 345.919,
      "text": "the next 6 months. Because after those",
      "source": "source_caption"
    },
    {
      "start": 345.919,
      "end": 348.24,
      "text": "six months are done, the model's pricing",
      "source": "source_caption"
    },
    {
      "start": 348.24,
      "end": 350.16,
      "text": "is actually, you know, a little bit more",
      "source": "source_caption"
    },
    {
      "start": 350.16,
      "end": 351.919,
      "text": "expensive. And now they show it at the",
      "source": "source_caption"
    },
    {
      "start": 351.919,
      "end": 354.72,
      "text": "bottom over here, you can see that after",
      "source": "source_caption"
    },
    {
      "start": 354.72,
      "end": 358.16,
      "text": "starting January of 2027, it will become",
      "source": "source_caption"
    },
    {
      "start": 358.16,
      "end": 362.479,
      "text": "$1.50 per input and $7.50 per output. So",
      "source": "source_caption"
    },
    {
      "start": 362.479,
      "end": 364.72,
      "text": "yeah, it's still cheap compared to the",
      "source": "source_caption"
    },
    {
      "start": 364.72,
      "end": 366.56,
      "text": "other frontier labs, but it's not as",
      "source": "source_caption"
    },
    {
      "start": 366.56,
      "end": 368.639,
      "text": "cheap as, for example, MU Spark when the",
      "source": "source_caption"
    },
    {
      "start": 368.639,
      "end": 371.199,
      "text": "pricing is updated or even the Deepsee",
      "source": "source_caption"
    },
    {
      "start": 371.199,
      "end": 373.28,
      "text": "version for Flash. But across these",
      "source": "source_caption"
    },
    {
      "start": 373.28,
      "end": 375.44,
      "text": "benchmarks, we can see that this model",
      "source": "source_caption"
    },
    {
      "start": 375.44,
      "end": 377.36,
      "text": "is better in many areas that they",
      "source": "source_caption"
    },
    {
      "start": 377.36,
      "end": 378.88,
      "text": "highlighted at the top. But then they",
      "source": "source_caption"
    },
    {
      "start": 378.88,
      "end": 381.199,
      "text": "also have some other areas like long",
      "source": "source_caption"
    },
    {
      "start": 381.199,
      "end": 382.96,
      "text": "video understanding, which this model",
      "source": "source_caption"
    },
    {
      "start": 382.96,
      "end": 387.199,
      "text": "excels at 85.4% versus 78.9%",
      "source": "source_caption"
    },
    {
      "start": 387.199,
      "end": 389.12,
      "text": "for the Terra model. And then the old",
      "source": "source_caption"
    },
    {
      "start": 389.12,
      "end": 390.88,
      "text": "model was also pretty good at that,",
      "source": "source_caption"
    },
    {
      "start": 390.88,
      "end": 392.55,
      "text": "84.2%.",
      "source": "source_caption"
    },
    {
      "start": 392.55,
      "end": 392.56,
      "text": "84.2%.",
      "source": "source_caption"
    },
    {
      "start": 392.56,
      "end": 394.8,
      "text": "Then long context performance the model",
      "source": "source_caption"
    },
    {
      "start": 394.8,
      "end": 399.44,
      "text": "is at 97% and this model the GPT 6 Terra",
      "source": "source_caption"
    },
    {
      "start": 399.44,
      "end": 402.16,
      "text": "one it sits at 93.5%.",
      "source": "source_caption"
    },
    {
      "start": 402.16,
      "end": 404.16,
      "text": "So yeah this model in summary it is",
      "source": "source_caption"
    },
    {
      "start": 404.16,
      "end": 406.479,
      "text": "better so it's not all negative but it's",
      "source": "source_caption"
    },
    {
      "start": 406.479,
      "end": 408.639,
      "text": "not all like you know that positive",
      "source": "source_caption"
    },
    {
      "start": 408.639,
      "end": 410.479,
      "text": "where you are super excited for Google",
      "source": "source_caption"
    },
    {
      "start": 410.479,
      "end": 412.24,
      "text": "Deep Mind because as I said they're",
      "source": "source_caption"
    },
    {
      "start": 412.24,
      "end": 413.84,
      "text": "probably still holding off their biggest",
      "source": "source_caption"
    },
    {
      "start": 413.84,
      "end": 416.319,
      "text": "release for Gemini 4 lineup. Now, one",
      "source": "source_caption"
    },
    {
      "start": 416.319,
      "end": 418.16,
      "text": "thing people might have missed in their",
      "source": "source_caption"
    },
    {
      "start": 418.16,
      "end": 419.759,
      "text": "charts because these charts sometimes",
      "source": "source_caption"
    },
    {
      "start": 419.759,
      "end": 422.8,
      "text": "are so messy to read and understand, but",
      "source": "source_caption"
    },
    {
      "start": 422.8,
      "end": 426.16,
      "text": "the 3.7 flash is worse than GPT 5.6",
      "source": "source_caption"
    },
    {
      "start": 426.16,
      "end": 428.16,
      "text": "Luna, which is the model over here,",
      "source": "source_caption"
    },
    {
      "start": 428.16,
      "end": 430.56,
      "text": "which is achieving a higher score on",
      "source": "source_caption"
    },
    {
      "start": 430.56,
      "end": 433.199,
      "text": "this benchmark deepware engineering for",
      "source": "source_caption"
    },
    {
      "start": 433.199,
      "end": 435.52,
      "text": "about three times the cost. So, yeah,",
      "source": "source_caption"
    },
    {
      "start": 435.52,
      "end": 437.039,
      "text": "this is kind of interesting because",
      "source": "source_caption"
    },
    {
      "start": 437.039,
      "end": 439.599,
      "text": "yeah, the cost for Gemini 3.7 Flash is a",
      "source": "source_caption"
    },
    {
      "start": 439.599,
      "end": 441.44,
      "text": "little bit more than what it looks like.",
      "source": "source_caption"
    },
    {
      "start": 441.44,
      "end": 443.36,
      "text": "Now, some people are a little bit upset",
      "source": "source_caption"
    },
    {
      "start": 443.36,
      "end": 445.12,
      "text": "and they're like, \"Oh, disgraceful.",
      "source": "source_caption"
    },
    {
      "start": 445.12,
      "end": 447.52,
      "text": "Google left out soul, opus, and fable",
      "source": "source_caption"
    },
    {
      "start": 447.52,
      "end": 449.68,
      "text": "because Google is incredibly behind. But",
      "source": "source_caption"
    },
    {
      "start": 449.68,
      "end": 451.039,
      "text": "I think one thing we got to remember,",
      "source": "source_caption"
    },
    {
      "start": 451.039,
      "end": 453.12,
      "text": "guys, is that this model is a flash",
      "source": "source_caption"
    },
    {
      "start": 453.12,
      "end": 455.919,
      "text": "model. It's not trying to be a pro model",
      "source": "source_caption"
    },
    {
      "start": 455.919,
      "end": 458.0,
      "text": "or it's not trying to compete with Opus",
      "source": "source_caption"
    },
    {
      "start": 458.0,
      "end": 460.639,
      "text": "or Soul or Fable 5 level models yet.",
      "source": "source_caption"
    },
    {
      "start": 460.639,
      "end": 462.8,
      "text": "That is probably going to be Gemini 4.",
      "source": "source_caption"
    },
    {
      "start": 462.8,
      "end": 464.88,
      "text": "So when Gemini 4 comes out, then I think",
      "source": "source_caption"
    },
    {
      "start": 464.88,
      "end": 467.199,
      "text": "it's okay for us to criticize them if",
      "source": "source_caption"
    },
    {
      "start": 467.199,
      "end": 469.44,
      "text": "they don't include Soul, Opus, and Fable",
      "source": "source_caption"
    },
    {
      "start": 469.44,
      "end": 471.199,
      "text": "in their benchmark charts because for",
      "source": "source_caption"
    },
    {
      "start": 471.199,
      "end": 473.199,
      "text": "now, I think what they have done is",
      "source": "source_caption"
    },
    {
      "start": 473.199,
      "end": 475.12,
      "text": "pretty accurate. One lab that I would",
      "source": "source_caption"
    },
    {
      "start": 475.12,
      "end": 476.639,
      "text": "have liked to see or one model for",
      "source": "source_caption"
    },
    {
      "start": 476.639,
      "end": 478.0,
      "text": "example I would have liked to see on",
      "source": "source_caption"
    },
    {
      "start": 478.0,
      "end": 479.84,
      "text": "that chart would be Deepseek version 4",
      "source": "source_caption"
    },
    {
      "start": 479.84,
      "end": 481.52,
      "text": "Flash because that would kind of spoil",
      "source": "source_caption"
    },
    {
      "start": 481.52,
      "end": 483.52,
      "text": "their release because that model is way",
      "source": "source_caption"
    },
    {
      "start": 483.52,
      "end": 486.319,
      "text": "cheaper compared to the Gemini 3.7 flash",
      "source": "source_caption"
    },
    {
      "start": 486.319,
      "end": 488.8,
      "text": "model lineup. Now this model jumped from",
      "source": "source_caption"
    },
    {
      "start": 488.8,
      "end": 491.199,
      "text": "number 19 to 8 on the web development",
      "source": "source_caption"
    },
    {
      "start": 491.199,
      "end": 493.599,
      "text": "area and we see it over here now and",
      "source": "source_caption"
    },
    {
      "start": 493.599,
      "end": 495.12,
      "text": "couple of models that are ahead of it",
      "source": "source_caption"
    },
    {
      "start": 495.12,
      "end": 498.16,
      "text": "are Opus 5 obviously Kim K3 Quinn 3.8",
      "source": "source_caption"
    },
    {
      "start": 498.16,
      "end": 501.68,
      "text": "Max Cloud Opus 5 Gro 4.6 6, which is a",
      "source": "source_caption"
    },
    {
      "start": 501.68,
      "end": 503.36,
      "text": "model that came yesterday, which was a",
      "source": "source_caption"
    },
    {
      "start": 503.36,
      "end": 506.72,
      "text": "big win for SpaceX, Fable 5, and 5.6.",
      "source": "source_caption"
    },
    {
      "start": 506.72,
      "end": 508.96,
      "text": "Soul. So, yeah, this model is trying to",
      "source": "source_caption"
    },
    {
      "start": 508.96,
      "end": 511.12,
      "text": "compete in the web development, but it's",
      "source": "source_caption"
    },
    {
      "start": 511.12,
      "end": 513.12,
      "text": "still behind all of these models, which",
      "source": "source_caption"
    },
    {
      "start": 513.12,
      "end": 515.12,
      "text": "is, you know, expected cuz it's a flash",
      "source": "source_caption"
    },
    {
      "start": 515.12,
      "end": 517.279,
      "text": "model. It's not really a pro model.",
      "source": "source_caption"
    },
    {
      "start": 517.279,
      "end": 519.599,
      "text": "Today, OpenAI has also launched a weight",
      "source": "source_caption"
    },
    {
      "start": 519.599,
      "end": 522.479,
      "text": "list for 5.6 so ultra fast mode. Now,",
      "source": "source_caption"
    },
    {
      "start": 522.479,
      "end": 524.959,
      "text": "this is possible because of the fast",
      "source": "source_caption"
    },
    {
      "start": 524.959,
      "end": 526.64,
      "text": "chips that they have access to now",
      "source": "source_caption"
    },
    {
      "start": 526.64,
      "end": 529.839,
      "text": "through Cabus and GPT 5.6 6o with this",
      "source": "source_caption"
    },
    {
      "start": 529.839,
      "end": 532.16,
      "text": "chip kind of running it is able to",
      "source": "source_caption"
    },
    {
      "start": 532.16,
      "end": 534.0,
      "text": "achieve an ultra fast mode that",
      "source": "source_caption"
    },
    {
      "start": 534.0,
      "end": 537.12,
      "text": "generates up to 750 output tokens per",
      "source": "source_caption"
    },
    {
      "start": 537.12,
      "end": 540.399,
      "text": "second which is about 14 times faster",
      "source": "source_caption"
    },
    {
      "start": 540.399,
      "end": 542.399,
      "text": "than the standard mode. So we're getting",
      "source": "source_caption"
    },
    {
      "start": 542.399,
      "end": 546.32,
      "text": "a really fast version of GPT 5.6 so now",
      "source": "source_caption"
    },
    {
      "start": 546.32,
      "end": 548.88,
      "text": "this is supposed to be used for live or",
      "source": "source_caption"
    },
    {
      "start": 548.88,
      "end": 550.72,
      "text": "near production workloads like you know",
      "source": "source_caption"
    },
    {
      "start": 550.72,
      "end": 553.279,
      "text": "real-time voice support commerce what",
      "source": "source_caption"
    },
    {
      "start": 553.279,
      "end": 555.839,
      "text": "this allows GPT 5.6 six soul to do is",
      "source": "source_caption"
    },
    {
      "start": 555.839,
      "end": 557.68,
      "text": "kind of be really fast in critical",
      "source": "source_caption"
    },
    {
      "start": 557.68,
      "end": 559.519,
      "text": "situations when people might be",
      "source": "source_caption"
    },
    {
      "start": 559.519,
      "end": 561.68,
      "text": "interacting with the AI agent like",
      "source": "source_caption"
    },
    {
      "start": 561.68,
      "end": 564.64,
      "text": "financial research security response",
      "source": "source_caption"
    },
    {
      "start": 564.64,
      "end": 566.48,
      "text": "support I think is going to be a big",
      "source": "source_caption"
    },
    {
      "start": 566.48,
      "end": 569.36,
      "text": "area where this new ultraast mode will",
      "source": "source_caption"
    },
    {
      "start": 569.36,
      "end": 572.0,
      "text": "be kind of implemented business and",
      "source": "source_caption"
    },
    {
      "start": 572.0,
      "end": 573.92,
      "text": "developer agents maybe yeah but I think",
      "source": "source_caption"
    },
    {
      "start": 573.92,
      "end": 575.6,
      "text": "like support or near production",
      "source": "source_caption"
    },
    {
      "start": 575.6,
      "end": 577.76,
      "text": "workloads like real-time voice I see",
      "source": "source_caption"
    },
    {
      "start": 577.76,
      "end": 580.16,
      "text": "this model really excelling at that and",
      "source": "source_caption"
    },
    {
      "start": 580.16,
      "end": 582.16,
      "text": "obviously this is still a weightless",
      "source": "source_caption"
    },
    {
      "start": 582.16,
      "end": 584.16,
      "text": "mode we don't know how many people are",
      "source": "source_caption"
    },
    {
      "start": 584.16,
      "end": 585.76,
      "text": "going to get access to this, how",
      "source": "source_caption"
    },
    {
      "start": 585.76,
      "end": 587.92,
      "text": "successful it is or what the pricing is.",
      "source": "source_caption"
    },
    {
      "start": 587.92,
      "end": 589.44,
      "text": "I don't know if the pricing has changed",
      "source": "source_caption"
    },
    {
      "start": 589.44,
      "end": 591.2,
      "text": "because there's no information in their",
      "source": "source_caption"
    },
    {
      "start": 591.2,
      "end": 594.0,
      "text": "actual, you know, blog post. But as I",
      "source": "source_caption"
    },
    {
      "start": 594.0,
      "end": 595.92,
      "text": "mentioned, couple of areas where they",
      "source": "source_caption"
    },
    {
      "start": 595.92,
      "end": 597.36,
      "text": "mentioned that this is going to be",
      "source": "source_caption"
    },
    {
      "start": 597.36,
      "end": 599.36,
      "text": "really important. Customer support and",
      "source": "source_caption"
    },
    {
      "start": 599.36,
      "end": 601.2,
      "text": "voice, commerce, live research and",
      "source": "source_caption"
    },
    {
      "start": 601.2,
      "end": 603.76,
      "text": "experimentation, financial research and",
      "source": "source_caption"
    },
    {
      "start": 603.76,
      "end": 605.839,
      "text": "security, incident response and",
      "source": "source_caption"
    },
    {
      "start": 605.839,
      "end": 607.76,
      "text": "reliability. But yeah, I think this",
      "source": "source_caption"
    },
    {
      "start": 607.76,
      "end": 609.68,
      "text": "partnership is going to be important for",
      "source": "source_caption"
    },
    {
      "start": 609.68,
      "end": 612.24,
      "text": "OpenAI going forward. But as I said, 14",
      "source": "source_caption"
    },
    {
      "start": 612.24,
      "end": 614.24,
      "text": "times the speed. What does that mean for",
      "source": "source_caption"
    },
    {
      "start": 614.24,
      "end": 616.8,
      "text": "cost? We don't know yet. But that's it",
      "source": "source_caption"
    },
    {
      "start": 616.8,
      "end": 618.56,
      "text": "for today's video. Make sure you guys",
      "source": "source_caption"
    },
    {
      "start": 618.56,
      "end": 620.48,
      "text": "are subscribed to the channel. Follow",
      "source": "source_caption"
    },
    {
      "start": 620.48,
      "end": 622.16,
      "text": "our new newsletter as well at",
      "source": "source_caption"
    },
    {
      "start": 622.16,
      "end": 624.389,
      "text": "universeai.behive.com",
      "source": "source_caption"
    },
    {
      "start": 624.389,
      "end": 624.399,
      "text": "universeai.behive.com",
      "source": "source_caption"
    },
    {
      "start": 624.399,
      "end": 626.079,
      "text": "as well as subscribe to the main channel",
      "source": "source_caption"
    },
    {
      "start": 626.079,
      "end": 628.56,
      "text": "World of AI and support us on X by",
      "source": "source_caption"
    },
    {
      "start": 628.56,
      "end": 630.8,
      "text": "following the Universe of AIZ as well.",
      "source": "source_caption"
    },
    {
      "start": 630.8,
      "end": 632.399,
      "text": "Until then, I'll see you guys in the",
      "source": "source_caption"
    },
    {
      "start": 632.399,
      "end": 634.56,
      "text": "next",
      "source": "source_caption"
    }
  ],
  "transcript_selection": {
    "schema": "videonote.transcript_selection.v1",
    "status": "SELECTED",
    "requested_language": "en",
    "duration": 633.16175,
    "reference_path": "/Users/sagawa/AI_WORK/video_notes/url/20260814_133848__Google_Shipped_Gemini_3.7_Flash_And_OpenAI_Made_GPT-5.6_14x_Faster/reference_subtitle.vtt",
    "reference_kind": "automatic_original",
    "reference_language": "en-orig",
    "selected_source": "youtube_auto_original",
    "decision_reason": "source_timing_retained_after_quality_gate",
    "whisper_rejected": false,
    "source_profile": {
      "segment_count": 285,
      "multilingual": false,
      "first_start": 0.24,
      "final_end": 634.56,
      "duration_coverage": 1.0,
      "maximum_gap_seconds": 0.24,
      "inverted_count": 0,
      "language": {
        "requested": "en",
        "matched": true,
        "latin": 8138,
        "han": 0,
        "kana": 0,
        "hangul": 0
      },
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "usable_full": true,
      "usable_partial": true
    },
    "whisper_profile": {
      "segment_count": 113,
      "multilingual": false,
      "first_start": 0.0,
      "final_end": 632.722,
      "duration_coverage": 0.999305,
      "maximum_gap_seconds": 1.0,
      "inverted_count": 0,
      "language": {
        "requested": "en",
        "matched": true,
        "latin": 8251,
        "han": 0,
        "kana": 0,
        "hangul": 0
      },
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "usable_full": true,
      "usable_partial": true
    },
    "comparison": {
      "window_seconds": 30.0,
      "compared_window_count": 22,
      "median_score": 0.9251,
      "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": []
    },
    "caption_timing_repair": {
      "method": "monotonic_multi_anchor_piecewise_interpolation",
      "applied": false,
      "anchor_count": 0,
      "anchors": [],
      "source_block_count": 54,
      "whisper_block_count": 46,
      "matched_block_count": 42,
      "case_normalization": {
        "applied": false,
        "changed_segment_count": 0
      },
      "source_text_preserved": true,
      "reason": "source_timing_retained_after_quality_gate"
    },
    "filled_source_profile": {
      "segment_count": 59,
      "multilingual": false,
      "first_start": 0.24,
      "final_end": 634.56,
      "duration_coverage": 1.0,
      "maximum_gap_seconds": 0.24,
      "inverted_count": 0,
      "language": {
        "requested": "en",
        "matched": true,
        "latin": 8138,
        "han": 0,
        "kana": 0,
        "hangul": 0
      },
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "usable_full": true,
      "usable_partial": true
    },
    "whisper_duration_guard": {
      "media_duration": 633.162,
      "rejected_segment_count": 0,
      "rejected_segments": []
    }
  }
}