{ "text": "So , looks like we just got flashed by Google DeepMind once again\nbecause they just dropped Gemini 3.7 Flash today.\nYes, their newest model, which is once again a Flash model,\nYes , their newest model , which is once again a Flash model ,\nis here .\nAnd this is only after 3 weeks since Gemini 3.6 Flash.\n6 Flash .\nAnd what they are saying is that Gemini 3.7 Flash is their most intelligent workhorse model.\n7 Flash is their most intelligent workhorse model .\nAnd this is coming 3 weeks after 3.6 Flash because of developer feedback and algorithmic innovations.\n6 six flash because of developer feedback and algorithmic innovations\n.\nAnd that is why they're saying the Gemini 3.7 Flash model is, you know, out with so much improvement cuz\n7 Flash model is , you know , out with so much improvement cuz\nif we take a look at the benchmarks , this model is way better\nthan the Gemini 3.6 Flash.\n6 Flash .\nBut I think this is also because we know internally that they're\nplanning on cancelling Gemini 3.5 Pro completely and working on Gemini 4 at the moment.\n5 Pro completely and working on Gemini 4 at the moment .\nSo maybe the improvements they made with the Pro model that they're\nnot going to be dropping anymore .\nMaybe they repackaged it as a flash model because this model is\nbetter than 3 .\nthe 3.5 Pro, it's quite likely that is the case.\nthe 3 .\n5 Pro , it's quite likely that is the case .\nWe'll just ignore that and put that into the side .\nWe don't know when we're getting a new Pro model from the Google\nwe will see that yes, it is stronger than the 3.6 Flash model.\nwill see that yes , it is stronger than the 3 .\n6 Flash model .\nFor example , when it comes to code quality , production code quality\nthe model is producing 43.6 versus the flash model 34.4.\n6 versus the flash model 34 .\n4 .\nAnd then Sonet 5 is at 42.7 and the Terra model is at 41.3.\n7 and the Terra model is at 41 .\n3 .\nSo , one thing to remember , since this is a flash model ,\nyou're not going to see them compare this to Opus 5 or any of the\nother stronger models .\nThe reason being cuz this is not their strongest tier .\nand the 3 .\n1 Pro model when they finally decide to upgrade it to Gemini 4\nPro , then we'll see it being compared to Opus 5 or GPT 5 .\n6 Soul .\nBut anyways , we can see that this model is an improvement from\n3 .\n6 Flash and I would say that's basically the biggest result or\nthe biggest update that we saw with the new model because it does\nnot really like change up things a lot is still like not becoming\nthe number one model or the number one flash model .\nEven on some benchmarks , Deepseek version 4 flash is actually\ncheaper and more intelligent than this model .\nBut let's just take a look at the benchmarks that they have published\n.\nOn Long Horizon software engineering , this model is a little bit\nbehind GPT 5 .\n6 Terra , which sits at 69 .\n6 and then 65 .\n3 is the Flash model .\nThe older Flash model 3 .\n6 sits at 48 .\n6 .\nSonic 5 sits at 53 .\n8 8 .\nAnd the new player that is finally being included on benchmarks\n, which even the Google DeepMind team is considering with their\nlaunch , is Muse Spark 1 .\n2 , which is sitting at 54 .\n9 .\nSo , welcome Meta to the benchmark charts because now we're starting\nto see it appear on more and more benchmarks .\nAnd if we also take a look at web development ,\nthis model is getting a ELO score of 1588 versus their old model\nis 1538 .\nSo , not a crazy difference , but compared to everything else out\nthere , this is number one .\nWhen I say everything else out there , once again ,\ncompared to all the mid tier models , this is beating all of them\nwhen it comes to web development .\nNow , this model is probably going to be used in enterprises a lot just because of Google's footprint in the enterprise space\nlot just because of Google's footprint in the enterprise space\n, but we are seeing this model achieve on the automation bench\n30 .\n4 .\nAnd GPT 5 .\n6 Terra sits at 23 .\n6 .\nSo , yes , this model is stronger than the other Flash models out\nthere , but as I said , they haven't included Deepseek version\nfor Flash because if they do , in some areas ,\nthat model is actually quite better than the 3 .\n7 Flash model .\nBefore we continue , if you're building AI agents or just messing\naround with them , Arcade is worth knowing about .\nIt's the runtime that lets your agent actually do things instead\nof just talking about them because that's the gap right now .\nThe models are smart enough .\nYour agent can figure out exactly what needs to happen in your\nemail , your Slack , your CRM .\nIt just can't go in and do it .\nAnd the reason isn't intelligence , it's permissions .\nSomething has to prove the agent is allowed to act on behalf of\na specific person in a specific account .\nThat's the messy part everyone runs into , and it's the partit\nactually handles for you .\nSo instead of your agent using one shared login for everybody ,\nit acts as whoever is actually signed in with exactly the access\nthat person has .\nIf they can't see something , the agent can't either .\nAnd you never have to touch any of that setup yourself .\nThen there's the tools .\nArcade has thousands of them already built for Gmail ,\nGoogle Drive , Slack , Notion , Salesforce , most of the apps people\nalready work in , and they are built specifically for AI to use\n.\nSo the agent gets it right the first time instead of guessing and\nfailing and retrying .\nIt also keeps a record of everything , what the agent did ,\nfor who and where , which matters a lot the moment other people\nstart using the thing you built .\nAnd it works with whatever you're already using .\nAny model , any framework , cloud , cursor , chat ,\nGPT , doesn't matter .\nSo you're not just giving an AI a list of tools and hoping it works\n.\nYou're giving it a place where it can safely take real actions\nin real apps .\nIt's free to start and the link is in the description .\nThank you once again for Arcade for sponsoring today's video .\nNow , let's get back into the video .\nNow , one thing to note is that the Gemini 3 .\n7 Flash model through the end of this year , so end of 2026 ,\nthey have a cheap pricing model that they're placing on the model\n.\n75 per 1 million input tokens and 3 .\n75 for 1 million output tokens .\nSo , it's a competitive price , but this is only for the next 6\nmonths .\nBecause after those six months are done , the model's pricing is\nactually , you know , a little bit more expensive .\nAnd now they show it at the bottom over here ,\nyou can see that after starting January of 2027 ,\nit will become 1 .\n50 per input and 7 .\n50 per output .\nSo yeah , it's still cheap compared to the other frontier labs\n, but it's not as cheap as , for example , MU Spark when the pricing\nis updated or even the Deepsee version for Flash .\nBut across these benchmarks , we can see that this model is better\nin many areas that they highlighted at the top .\nBut then they also have some other areas like long video understanding\n, which this model excels at 85 .\n4 versus 78 .\n9 for the Terra model .\nAnd then the old model was also pretty good at that ,\n84 .\n2 .\nThen long context performance the model is at 97 and this model\nthe GPT 6 Terra one it sits at 93 .\n5.\nSo yeah this model in summary it is better so it's not all negative\nbut it's not all like you know that positive where you are super\nexcited for Google Deep Mind because as I said they're probably\nstill holding off their biggest release for Gemini 4 lineup .\nNow , one thing people might have missed in their charts because\nthese charts sometimes are so messy to read and understand ,\n3.\n7 flash is worse than GPT 5.\n6 Luna , which is the model over here , which is achieving a higher\nscore on this benchmark deepware engineering for about three times\nthe cost .\nSo , yeah , this is kind of interesting because yeah ,\nthe cost for Gemini 3.\n7 Flash is a little bit more than what it looks like .\nNow , some people are a little bit upset and they're like ,\nOh , disgraceful .\nGoogle left out soul , opus , and fable because Google is incredibly\nbehind .\nBut I think one thing we got to remember , guys ,\nis that this model is a flash model .\nIt's not trying to be a pro model or it's not trying to compete\nThat is probably going to be Gemini 4 .\nThat is probably going to be Gemini 4 .\nSo when Gemini 4 comes out , then I think it's okay for us to criticize\nthem if they don't include Soul , Opus , and Fable in their benchmark\ncharts because for now , I think what they have done is pretty\naccurate .\nOne lab that I would have liked to see or one model for example\nI would have liked to see on that chart would be Deepseek version\n4 Flash because that would kind of spoil their release because\nthat model is way cheaper compared to the Gemini 3.\n7 flash model lineup .\nNow this model jumped from number 19 to 8 on the web development\narea and we see it over here now and couple of models that are\nahead of it are Opus 5 obviously Kim K3 Quinn 3 .\n8 Max Cloud Opus 5 Gro 4 .\n6 6 , which is a model that came yesterday , which was a big win\nfor SpaceX, Fable 5, and 5.6.\n6.\nSol.\nSo , yeah , this model is trying to compete in the web development\n, but it's still behind all of these models , which is ,\nyou know , expected cuz it's a flash model .\nIt's not really a pro model .\nToday, OpenAI has also launched a weight list for 5.6 so ultra fast mode.\n6 so ultra fast mode .\naccess to now through Cerebras and GPT 5.\n6 6o with this chip kind of running it is able to achieve an ultra\n6 6o with this chip kind of running it is able to achieve an ultra\nfast mode that generates up to 750 output tokens per second which\nis about 14 times faster than the standard mode .\nSo we're getting a really fast version of GPT 5 .\n6 so now this is supposed to be used for live or near production\nworkloads like you know real time voice support commerce what this\n6 six sol to do is kind of be really fast in critical situations\n6 six soul to do is kind of be really fast in critical situations\nwhen people might be interacting with the AI agent like financial\nresearch security response support I think is going to be a big\narea where this new ultrafast mode will be kind of implemented business\nand developer agents maybe yeah but I think like support or near\nproduction workloads like real time voice I see this model really\nexcelling at that and obviously this is still a weightless mode\nwe don't know how many people are going to get access to this ,\nhow successful it is or what the pricing is .\nI don't know if the pricing has changed because there's no information\nin their actual , you know , blog post .\nBut as I mentioned , couple of areas where they mentioned that\nthis is going to be really important .\nCustomer support and voice , commerce , live research and experimentation\n, financial research and security , incident response and reliability\n.\nBut yeah , I think this partnership is going to be important for\nOpenAI going forward .\nBut as I said , 14 times the speed .\nWhat does that mean for cost ?\nWe don't know yet .\nBut that's it for today's video .\nMake sure you guys are subscribed to the channel .\nFollow our new newsletter as well at universe-of-ai.beehiiv.com as well as subscribe to the main channel World of AI and support\nbehive .\ncom as well as subscribe to the main channel World of AI and support\nus on X by following the Universe of AI as well .\nUntil then , I'll see you guys in the next", "language": "en", "duration": 633.16175, "segments": [ { "start": 0.0, "end": 4.41, "text": "So , looks like we just got flashed by Google DeepMind once again", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 4.42, "end": 6.412, "text": "because they just dropped Gemini 3.7 Flash today.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "because they just dropped Gemini 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 6.422, "end": 7.705, "text": "Yes, their newest model, which is once again a Flash model,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "7 Flash today .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 7.715, "end": 11.162, "text": "Yes , their newest model , which is once again a Flash model ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 11.172, "end": 12.046, "text": "is here .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 12.056, "end": 15.473, "text": "And this is only after 3 weeks since Gemini 3.6 Flash.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "And this is only after 3 weeks since Gemini 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 15.483, "end": 16.544, "text": "6 Flash .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 16.554, "end": 19.034, "text": "And what they are saying is that Gemini 3.7 Flash is their most intelligent workhorse model.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "And what they are saying is that Gemini 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 19.044, "end": 22.649, "text": "7 Flash is their most intelligent workhorse model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 22.659, "end": 24.772, "text": "And this is coming 3 weeks after 3.6 Flash because of developer feedback and algorithmic innovations.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "And this is coming 3 weeks after 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 24.782, "end": 29.103, "text": "6 six flash because of developer feedback and algorithmic innovations", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 29.113, "end": 29.786, "text": ".", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 29.796, "end": 31.697, "text": "And that is why they're saying the Gemini 3.7 Flash model is, you know, out with so much improvement cuz", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "And that is why they're saying the Gemini 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 31.707, "end": 35.352, "text": "7 Flash model is , you know , out with so much improvement cuz", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 35.362, "end": 38.35, "text": "if we take a look at the benchmarks , this model is way better", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 38.36, "end": 39.552, "text": "than the Gemini 3.6 Flash.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "than the Gemini 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 39.562, "end": 40.622, "text": "6 Flash .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 40.632, "end": 43.836, "text": "But I think this is also because we know internally that they're", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 43.846, "end": 45.717, "text": "planning on cancelling Gemini 3.5 Pro completely and working on Gemini 4 at the moment.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "planning on cancelelling Gemini 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 45.727, "end": 49.647, "text": "5 Pro completely and working on Gemini 4 at the moment .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 49.657, "end": 52.595, "text": "So maybe the improvements they made with the Pro model that they're", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 52.605, "end": 54.432, "text": "not going to be dropping anymore .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 54.442, "end": 58.36, "text": "Maybe they repackaged it as a flash model because this model is", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 58.37, "end": 59.537, "text": "better than 3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 59.547, "end": 63.495, "text": "the 3.5 Pro, it's quite likely that is the case.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "6 and if the timeline is three weeks plus they're planning on cancelling", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 63.505, "end": 64.083, "text": "the 3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 64.093, "end": 66.949, "text": "5 Pro , it's quite likely that is the case .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 66.959, "end": 68.958, "text": "We'll just ignore that and put that into the side .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 68.968, "end": 71.45, "text": "We don't know when we're getting a new Pro model from the Google", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 71.46, "end": 75.058, "text": "we will see that yes, it is stronger than the 3.6 Flash model.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "Deep Mind team , but this model across many of the benchmarks we", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 75.068, "end": 78.319, "text": "will see that yes , it is stronger than the 3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 78.329, "end": 79.628, "text": "6 Flash model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 79.638, "end": 82.654, "text": "For example , when it comes to code quality , production code quality", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 82.664, "end": 85.128, "text": "the model is producing 43.6 versus the flash model 34.4.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": ", the model is producing 43 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 85.138, "end": 88.11, "text": "6 versus the flash model 34 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 88.12, "end": 89.213, "text": "4 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 89.223, "end": 92.352, "text": "And then Sonet 5 is at 42.7 and the Terra model is at 41.3.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "And then Sonet 5 is at 42 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 92.362, "end": 95.57, "text": "7 and the Terra model is at 41 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 95.58, "end": 96.362, "text": "3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 96.372, "end": 99.346, "text": "So , one thing to remember , since this is a flash model ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 99.356, "end": 102.081, "text": "you're not going to see them compare this to Opus 5 or any of the", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 102.091, "end": 103.572, "text": "other stronger models .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 103.582, "end": 106.775, "text": "The reason being cuz this is not their strongest tier .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 106.785, "end": 107.65, "text": "and the 3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 107.66, "end": 111.39, "text": "1 Pro model when they finally decide to upgrade it to Gemini 4", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 111.4, "end": 116.013, "text": "Pro , then we'll see it being compared to Opus 5 or GPT 5 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 116.023, "end": 116.842, "text": "6 Soul .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 116.852, "end": 119.728, "text": "But anyways , we can see that this model is an improvement from", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 119.738, "end": 120.209, "text": "3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 120.219, "end": 124.699, "text": "6 Flash and I would say that's basically the biggest result or", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 124.709, "end": 127.986, "text": "the biggest update that we saw with the new model because it does", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 127.996, "end": 131.391, "text": "not really like change up things a lot is still like not becoming", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 131.401, "end": 134.208, "text": "the number one model or the number one flash model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 134.218, "end": 138.547, "text": "Even on some benchmarks , Deepseek version 4 flash is actually", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 138.557, "end": 141.54, "text": "cheaper and more intelligent than this model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 141.55, "end": 144.249, "text": "But let's just take a look at the benchmarks that they have published", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 144.259, "end": 144.613, "text": ".", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 144.623, "end": 148.17, "text": "On Long Horizon software engineering , this model is a little bit", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 148.18, "end": 149.583, "text": "behind GPT 5 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 149.593, "end": 152.039, "text": "6 Terra , which sits at 69 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 152.049, "end": 153.991, "text": "6 and then 65 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 154.001, "end": 155.751, "text": "3 is the Flash model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 155.761, "end": 157.686, "text": "The older Flash model 3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 157.696, "end": 159.29, "text": "6 sits at 48 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 159.3, "end": 160.23, "text": "6 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 160.24, "end": 161.987, "text": "Sonic 5 sits at 53 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 161.997, "end": 162.941, "text": "8 8 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 162.951, "end": 166.38, "text": "And the new player that is finally being included on benchmarks", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 166.39, "end": 169.818, "text": ", which even the Google DeepMind team is considering with their", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 169.828, "end": 171.508, "text": "launch , is Muse Spark 1 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 171.518, "end": 173.699, "text": "2 , which is sitting at 54 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 173.709, "end": 174.544, "text": "9 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 174.554, "end": 178.698, "text": "So , welcome Meta to the benchmark charts because now we're starting", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 178.708, "end": 181.254, "text": "to see it appear on more and more benchmarks .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 181.264, "end": 183.571, "text": "And if we also take a look at web development ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 183.581, "end": 187.699, "text": "this model is getting a ELO score of 1588 versus their old model", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 187.709, "end": 188.663, "text": "is 1538 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 188.673, "end": 191.345, "text": "So , not a crazy difference , but compared to everything else out", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 191.355, "end": 192.726, "text": "there , this is number one .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 192.736, "end": 194.757, "text": "When I say everything else out there , once again ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 194.767, "end": 198.116, "text": "compared to all the mid tier models , this is beating all of them", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 198.126, "end": 199.891, "text": "when it comes to web development .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 199.901, "end": 202.979, "text": "Now , this model is probably going to be used in enterprises a lot just because of Google's footprint in the enterprise space", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "Now , this model is probably going to be used in enterprises a", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 202.989, "end": 205.758, "text": "lot just because of Google's footprint in the enterprise space", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 205.768, "end": 209.462, "text": ", but we are seeing this model achieve on the automation bench", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 209.472, "end": 210.19, "text": "30 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 210.2, "end": 210.958, "text": "4 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 210.968, "end": 212.111, "text": "And GPT 5 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 212.121, "end": 214.01, "text": "6 Terra sits at 23 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 214.02, "end": 214.89, "text": "6 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 214.9, "end": 218.56, "text": "So , yes , this model is stronger than the other Flash models out", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 218.57, "end": 221.588, "text": "there , but as I said , they haven't included Deepseek version", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 221.598, "end": 224.142, "text": "for Flash because if they do , in some areas ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 224.152, "end": 226.458, "text": "that model is actually quite better than the 3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 226.468, "end": 228.055, "text": "7 Flash model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 228.065, "end": 231.409, "text": "Before we continue , if you're building AI agents or just messing", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 231.419, "end": 234.285, "text": "around with them , Arcade is worth knowing about .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 234.295, "end": 237.318, "text": "It's the runtime that lets your agent actually do things instead", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 237.328, "end": 240.433, "text": "of just talking about them because that's the gap right now .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 240.443, "end": 242.109, "text": "The models are smart enough .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 242.119, "end": 245.23, "text": "Your agent can figure out exactly what needs to happen in your", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 245.24, "end": 247.309, "text": "email , your Slack , your CRM .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 247.319, "end": 249.389, "text": "It just can't go in and do it .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 249.399, "end": 252.431, "text": "And the reason isn't intelligence , it's permissions .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 252.441, "end": 255.39, "text": "Something has to prove the agent is allowed to act on behalf of", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 255.4, "end": 258.35, "text": "a specific person in a specific account .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 258.36, "end": 262.032, "text": "That's the messy part everyone runs into , and it's the partit", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 262.042, "end": 263.393, "text": "actually handles for you .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 263.403, "end": 266.835, "text": "So instead of your agent using one shared login for everybody ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 266.845, "end": 270.516, "text": "it acts as whoever is actually signed in with exactly the access", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 270.526, "end": 271.878, "text": "that person has .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 271.888, "end": 274.76, "text": "If they can't see something , the agent can't either .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 274.77, "end": 277.482, "text": "And you never have to touch any of that setup yourself .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 277.492, "end": 278.843, "text": "Then there's the tools .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 278.853, "end": 281.725, "text": "Arcade has thousands of them already built for Gmail ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 281.735, "end": 285.728, "text": "Google Drive , Slack , Notion , Salesforce , most of the apps people", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 285.738, "end": 289.09, "text": "already work in , and they are built specifically for AI to use", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 289.1, "end": 289.33, "text": ".", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 289.34, "end": 292.371, "text": "So the agent gets it right the first time instead of guessing and", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 292.381, "end": 293.733, "text": "failing and retrying .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 293.743, "end": 296.855, "text": "It also keeps a record of everything , what the agent did ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 296.865, "end": 300.138, "text": "for who and where , which matters a lot the moment other people", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 300.148, "end": 301.739, "text": "start using the thing you built .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 301.749, "end": 304.14, "text": "And it works with whatever you're already using .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 304.15, "end": 307.182, "text": "Any model , any framework , cloud , cursor , chat ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 307.192, "end": 308.943, "text": "GPT , doesn't matter .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 308.953, "end": 311.986, "text": "So you're not just giving an AI a list of tools and hoping it works", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 311.996, "end": 312.226, "text": ".", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 312.236, "end": 315.188, "text": "You're giving it a place where it can safely take real actions", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 315.198, "end": 316.469, "text": "in real apps .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 316.479, "end": 319.03, "text": "It's free to start and the link is in the description .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 319.04, "end": 322.066, "text": "Thank you once again for Arcade for sponsoring today's video .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 322.076, "end": 324.378, "text": "Now , let's get back into the video .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 324.388, "end": 327.178, "text": "Now , one thing to note is that the Gemini 3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 327.188, "end": 331.007, "text": "7 Flash model through the end of this year , so end of 2026 ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 331.017, "end": 334.283, "text": "they have a cheap pricing model that they're placing on the model", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 334.293, "end": 334.63, "text": ".", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 334.64, "end": 338.786, "text": "75 per 1 million input tokens and 3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 338.796, "end": 340.822, "text": "75 for 1 million output tokens .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 340.832, "end": 344.266, "text": "So , it's a competitive price , but this is only for the next 6", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 344.276, "end": 344.899, "text": "months .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 344.909, "end": 348.305, "text": "Because after those six months are done , the model's pricing is", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 348.315, "end": 350.528, "text": "actually , you know , a little bit more expensive .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 350.538, "end": 352.922, "text": "And now they show it at the bottom over here ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 352.932, "end": 356.99, "text": "you can see that after starting January of 2027 ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 357.0, "end": 358.763, "text": "it will become 1 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 358.773, "end": 360.728, "text": "50 per input and 7 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 360.738, "end": 361.947, "text": "50 per output .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 361.957, "end": 365.605, "text": "So yeah , it's still cheap compared to the other frontier labs", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 365.615, "end": 368.896, "text": ", but it's not as cheap as , for example , MU Spark when the pricing", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 368.906, "end": 372.352, "text": "is updated or even the Deepsee version for Flash .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 372.362, "end": 375.879, "text": "But across these benchmarks , we can see that this model is better", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 375.889, "end": 378.283, "text": "in many areas that they highlighted at the top .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 378.293, "end": 381.768, "text": "But then they also have some other areas like long video understanding", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 381.778, "end": 384.452, "text": ", which this model excels at 85 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 384.462, "end": 386.501, "text": "4 versus 78 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 386.511, "end": 388.298, "text": "9 for the Terra model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 388.308, "end": 391.897, "text": "And then the old model was also pretty good at that ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 391.907, "end": 392.176, "text": "84 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 392.186, "end": 392.465, "text": "2 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 392.475, "end": 397.432, "text": "Then long context performance the model is at 97 and this model", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 397.442, "end": 401.273, "text": "the GPT 6 Terra one it sits at 93 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 401.283, "end": 402.08, "text": "5.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "5 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 402.09, "end": 405.84, "text": "So yeah this model in summary it is better so it's not all negative", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 405.85, "end": 409.6, "text": "but it's not all like you know that positive where you are super", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 409.61, "end": 412.45, "text": "excited for Google Deep Mind because as I said they're probably", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 412.46, "end": 415.689, "text": "still holding off their biggest release for Gemini 4 lineup .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 415.699, "end": 418.635, "text": "Now , one thing people might have missed in their charts because", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 418.645, "end": 422.217, "text": "these charts sometimes are so messy to read and understand ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 422.227, "end": 423.412, "text": "3.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "but the 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 423.422, "end": 425.822, "text": "7 flash is worse than GPT 5.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "7 flash is worse than GPT 5 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 425.832, "end": 429.464, "text": "6 Luna , which is the model over here , which is achieving a higher", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 429.474, "end": 433.841, "text": "score on this benchmark deepware engineering for about three times", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 433.851, "end": 434.809, "text": "the cost .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 434.819, "end": 436.992, "text": "So , yeah , this is kind of interesting because yeah ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 437.002, "end": 438.34, "text": "the cost for Gemini 3.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "the cost for Gemini 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 438.35, "end": 441.276, "text": "7 Flash is a little bit more than what it looks like .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 441.286, "end": 443.781, "text": "Now , some people are a little bit upset and they're like ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 443.791, "end": 444.994, "text": "Oh , disgraceful .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 445.004, "end": 448.734, "text": "Google left out soul , opus , and fable because Google is incredibly", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 448.744, "end": 449.447, "text": "behind .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 449.457, "end": 451.115, "text": "But I think one thing we got to remember , guys ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 451.125, "end": 453.336, "text": "is that this model is a flash model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 453.346, "end": 456.986, "text": "It's not trying to be a pro model or it's not trying to compete", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 456.996, "end": 460.458, "text": "That is probably going to be Gemini 4 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "with Opus or Soul or Fable 5 level models yet .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 460.468, "end": 462.651, "text": "That is probably going to be Gemini 4 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 462.661, "end": 466.303, "text": "So when Gemini 4 comes out , then I think it's okay for us to criticize", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 466.313, "end": 470.156, "text": "them if they don't include Soul , Opus , and Fable in their benchmark", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 470.166, "end": 473.35, "text": "charts because for now , I think what they have done is pretty", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 473.36, "end": 474.069, "text": "accurate .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 474.079, "end": 476.864, "text": "One lab that I would have liked to see or one model for example", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 476.874, "end": 479.478, "text": "I would have liked to see on that chart would be Deepseek version", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 479.488, "end": 482.134, "text": "4 Flash because that would kind of spoil their release because", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 482.144, "end": 485.521, "text": "that model is way cheaper compared to the Gemini 3.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "that model is way cheaper compared to the Gemini 3 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 485.531, "end": 487.058, "text": "7 flash model lineup .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 487.068, "end": 490.807, "text": "Now this model jumped from number 19 to 8 on the web development", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 490.817, "end": 494.224, "text": "area and we see it over here now and couple of models that are", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 494.234, "end": 497.751, "text": "ahead of it are Opus 5 obviously Kim K3 Quinn 3 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 497.761, "end": 500.384, "text": "8 Max Cloud Opus 5 Gro 4 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 500.394, "end": 503.353, "text": "6 6 , which is a model that came yesterday , which was a big win", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 503.363, "end": 505.9, "text": "for SpaceX, Fable 5, and 5.6.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "for SpaceX , Fable 5 , and 5 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 505.91, "end": 506.352, "text": "6.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "6 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 506.362, "end": 507.002, "text": "Sol.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "Soul .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 507.012, "end": 509.966, "text": "So , yeah , this model is trying to compete in the web development", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 509.976, "end": 513.094, "text": ", but it's still behind all of these models , which is ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 513.104, "end": 515.305, "text": "you know , expected cuz it's a flash model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 515.315, "end": 517.111, "text": "It's not really a pro model .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 517.121, "end": 520.296, "text": "Today, OpenAI has also launched a weight list for 5.6 so ultra fast mode.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "Today , OpenAI has also launched a weight list for 5 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 520.306, "end": 522.036, "text": "6 so ultra fast mode .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 522.046, "end": 525.623, "text": "access to now through Cerebras and GPT 5.", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "Now , this is possible because of the fast chips that they have", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 525.633, "end": 528.49, "text": "6 6o with this chip kind of running it is able to achieve an ultra", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "access to now through Cabus and GPT 5 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 528.5, "end": 532.774, "text": "6 6o with this chip kind of running it is able to achieve an ultra", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 532.784, "end": 537.861, "text": "fast mode that generates up to 750 output tokens per second which", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 537.871, "end": 541.494, "text": "is about 14 times faster than the standard mode .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 541.504, "end": 544.812, "text": "So we're getting a really fast version of GPT 5 .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 544.822, "end": 549.505, "text": "6 so now this is supposed to be used for live or near production", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 549.515, "end": 553.292, "text": "workloads like you know real time voice support commerce what this", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 553.302, "end": 554.581, "text": "6 six sol to do is kind of be really fast in critical situations", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "allows GPT 5 .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 554.591, "end": 558.277, "text": "6 six soul to do is kind of be really fast in critical situations", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 558.287, "end": 562.135, "text": "when people might be interacting with the AI agent like financial", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 562.145, "end": 566.395, "text": "research security response support I think is going to be a big", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 566.405, "end": 571.508, "text": "area where this new ultrafast mode will be kind of implemented business", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "area where this new ultraast mode will be kind of implemented business", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 571.518, "end": 575.098, "text": "and developer agents maybe yeah but I think like support or near", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 575.108, "end": 578.527, "text": "production workloads like real time voice I see this model really", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 578.537, "end": 582.378, "text": "excelling at that and obviously this is still a weightless mode", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 582.388, "end": 585.323, "text": "we don't know how many people are going to get access to this ,", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 585.333, "end": 587.791, "text": "how successful it is or what the pricing is .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 587.801, "end": 590.497, "text": "I don't know if the pricing has changed because there's no information", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 590.507, "end": 593.283, "text": "in their actual , you know , blog post .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 593.293, "end": 596.396, "text": "But as I mentioned , couple of areas where they mentioned that", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 596.406, "end": 598.233, "text": "this is going to be really important .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 598.243, "end": 601.669, "text": "Customer support and voice , commerce , live research and experimentation", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 601.679, "end": 606.182, "text": ", financial research and security , incident response and reliability", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 606.192, "end": 606.701, "text": ".", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 606.711, "end": 609.514, "text": "But yeah , I think this partnership is going to be important for", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 609.524, "end": 611.044, "text": "OpenAI going forward .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 611.054, "end": 612.978, "text": "But as I said , 14 times the speed .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 612.988, "end": 614.589, "text": "What does that mean for cost ?", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 614.599, "end": 616.12, "text": "We don't know yet .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 616.13, "end": 617.812, "text": "But that's it for today's video .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 617.822, "end": 620.068, "text": "Make sure you guys are subscribed to the channel .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 620.078, "end": 623.626, "text": "Follow our new newsletter as well at universe-of-ai.beehiiv.com as well as subscribe to the main channel World of AI and support", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "Follow our new newsletter as well at universeai .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 623.636, "end": 624.014, "text": "behive .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 624.024, "end": 626.911, "text": "com as well as subscribe to the main channel World of AI and support", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" }, { "start": 626.921, "end": 629.396, "text": "us on X by following the Universe of AI as well .", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow", "text_original": "us on X by following the Universe of AIZ as well .", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 629.406, "end": 632.401, "text": "Until then , I'll see you guys in the next", "source": "source_caption", "timing_source": "vad_whisper_monotonic_reflow" } ], "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": [] } }, "caption_alignment": { "method": "vad_whisper_monotonic_reflow", "applied": true, "source_text_authority": "preserved" }, "presegmentation_correction": { "applied": true, "correction_count": 29 } }