1 00:00:00,000 --> 00:00:05,740 So DeepSeek version 4 Pro is officially out today. Now you might be confused because this model was 2 00:00:05,740 --> 00:00:10,680 technically out but it was in general availability but this is the official release meaning that they 3 00:00:10,680 --> 00:00:15,200 have trained the model a little bit more and produced a stronger version of the model that 4 00:00:15,200 --> 00:00:19,900 is available today. Now if I were to summarize today's release in one simple sentence it is that 5 00:00:19,900 --> 00:00:25,000 the price to performance is becoming a very very important thing because not only do we have a new 6 00:00:25,000 --> 00:00:30,660 model from the DeepSeek team, the SpaceX team also dropped Grok 4.6 and both of these models 7 00:00:30,660 --> 00:00:36,520 are competing with the best frontier labs at a fraction of their cost. Let's start with Grok 4.6 8 00:00:36,520 --> 00:00:41,180 and one thing I'm going to say is that I'm genuinely surprised by the SpaceX team. I'm not 9 00:00:41,180 --> 00:00:45,860 trying to glaze them or Elon Musk or anything. I was just very critical of this lab. In 2025, 10 00:00:46,240 --> 00:00:51,380 they dropped Grok 4 and other models like that but I wasn't really you know mind blown with their 11 00:00:51,380 --> 00:00:56,240 performance because they're pretty they're pretty subpar compared to any of the other labs out there 12 00:00:56,240 --> 00:01:01,680 but in 2026 it looks like things have kind of changed a little because grok 4.5 was quite 13 00:01:01,680 --> 00:01:08,280 competitive based off what it cost and grok 4.6 is actually not too bad if we take a look at the 14 00:01:08,280 --> 00:01:13,300 benchmarks for example if we start with the artificial analysis intelligence index this 15 00:01:13,300 --> 00:01:20,080 model achieves a 61 and fable 5 is at 62 now what's really important to remember is that once again 16 00:01:20,080 --> 00:01:25,460 this model is quite cheap compared to Fable 5. This is about $2 per million input tokens 17 00:01:25,460 --> 00:01:31,900 and $6 per million output tokens. And Fable 5 sits at $10 per million input tokens and $50 18 00:01:31,900 --> 00:01:36,700 per million output tokens. And this is why you start to appreciate this release a little bit 19 00:01:36,700 --> 00:01:42,240 more. It might not beat the performance of the best models. It's matching them. Even GPT 5.6 20 00:01:42,240 --> 00:01:47,540 sold, which is a pretty capable model, and this is set at max on the artificial analysis index. 21 00:01:47,540 --> 00:01:54,660 this model achieves 61 and grok 4.6 61 so it ties it and it's much cheaper and then even on all of 22 00:01:54,660 --> 00:02:00,420 these other benchmarks for example the gdp valve one it actually beats fable 5 which is at 1741 23 00:02:00,420 --> 00:02:06,900 and then gpt 5.6 it's at 1728 i'm not sure why they didn't choose opus 5 as well but i guess 24 00:02:06,900 --> 00:02:11,460 they wanted to choose the quote-unquote strongest model lineup from each lab and they chose fable 25 00:02:11,460 --> 00:02:16,140 5 for anthropic which is fair and then deep software engineering one which is a critical 26 00:02:16,140 --> 00:02:23,720 benchmark this model doesn't beat fable 5 or gpt 5.6 soul but it gets close to it it's 65.9 and 27 00:02:23,720 --> 00:02:29,040 fable 5 sits at 70 but if you're getting results that are pretty close and the model is five times 28 00:02:29,040 --> 00:02:34,080 cheaper i wouldn't be too disappointed with this result and then same thing with cursor bench 3.2 29 00:02:34,080 --> 00:02:42,240 the model achieves 69.9 funny number and fable 5 sits at 70.5 so once again closer and grok 4.6 30 00:02:42,240 --> 00:02:48,200 beats GPT 5.6 soul. The same thing with the Frontier Code, it gets close to Fable 5 beats 31 00:02:48,200 --> 00:02:53,940 GPT 5.6 soul. So yeah, this model is actually available in Cursor. So the partnership with 32 00:02:53,940 --> 00:02:58,960 Cursor or I guess the acquisition has really helped SpaceX make some strides in the AI space 33 00:02:58,960 --> 00:03:04,120 this year. And Grok build is something that, you know, maybe not a lot of us have been using so 34 00:03:04,120 --> 00:03:09,480 far, but it's probably going to be another platform like Codex or Cloud Code that we start to use. 35 00:03:09,480 --> 00:03:14,580 but obviously cursor is quite strong as well so you have options available for you to use them in 36 00:03:14,580 --> 00:03:20,240 both and one thing to note is that they're offering two times usage inside grok build and cursor for 37 00:03:20,240 --> 00:03:25,320 the first week so if you just want to try it out see what you feel about it then you know it might 38 00:03:25,320 --> 00:03:29,760 be worth trying it out right now because you get double the usage in the first week now if we were 39 00:03:29,760 --> 00:03:34,520 to take a look at some of the outputs that people have been generating with grok 4.6 what we're 40 00:03:34,520 --> 00:03:40,180 looking at right now is a Falcon 9 booster return sequence simulation and this was done in a single 41 00:03:40,180 --> 00:03:47,620 HTML file and as I said if you expected to get this type of output from Grok in 2025 you would 42 00:03:47,620 --> 00:03:52,140 be kind of surprised because you wouldn't expect something like this to be generated with Grok but 43 00:03:52,140 --> 00:03:55,960 now it looks like we have to start taking the Grok team a little bit more serious because 44 00:03:55,960 --> 00:04:01,160 this output is quite competitive and obviously we're just looking at a simulation and we're just 45 00:04:01,160 --> 00:04:05,940 basing it off of a visual representation but if the model is able to produce something like this 46 00:04:05,940 --> 00:04:11,340 consistently then I would expect a lot of people to start adopting Grok because number one it is 47 00:04:11,340 --> 00:04:15,660 cheaper than the other labs at least at the moment because we don't know if this pricing strategy is 48 00:04:15,660 --> 00:04:20,360 going to be sustainable for the SpaceX team in the long run but at least for now their models are 49 00:04:20,360 --> 00:04:25,280 definitely cheaper compared to the others and as I mentioned this model excelled at the artificial 50 00:04:25,280 --> 00:04:32,580 analysis index. This model jumped to probably number four model. It's tied to GPT 5.6. It's 51 00:04:32,580 --> 00:04:36,560 pretty much similar. So you could say number three as well. But the models before that are 52 00:04:36,560 --> 00:04:43,540 Fable 5 and Opus 5, which are only above the model by about a 1% or a 2% difference. So yeah, 53 00:04:43,580 --> 00:04:48,640 even on this intelligence index, which if you're not familiar with, has nine evaluations. So on 54 00:04:48,640 --> 00:04:54,220 all of these evaluations, it's kind of matching almost Fable 5 performance, which is crazy to see. 55 00:04:54,220 --> 00:04:59,760 Before we continue, we just launched the Universe of AI newsletter. If you want to stay on top of 56 00:04:59,760 --> 00:05:05,060 AI news without having to hunt for it, link is in the description. Don't miss out. And what you see 57 00:05:05,060 --> 00:05:10,980 on screen right now is a racing game that Grok 4.6 built. And based off of this post, the model took 58 00:05:10,980 --> 00:05:15,780 about one minute and it was a five word prompt, which was a create a simple racing game in HTML. 59 00:05:16,280 --> 00:05:21,200 So if you're able to generate something like this easily using Grok 4.6, I think a lot of people 60 00:05:21,200 --> 00:05:27,160 will be happy. And this is a more detailed analysis of what it costs to run GPT 5.6 on 61 00:05:27,160 --> 00:05:32,120 the artificial analysis index and what it produced, meaning the output. Both of these models, if you 62 00:05:32,120 --> 00:05:38,120 remember, scored 61 on the artificial analysis index. Now to run the whole test with GPT 5.6, 63 00:05:38,480 --> 00:05:45,500 it costs about 2.8k. And then with Grok 4.6, it costs about 1.1k-ish. And this tells you that 64 00:05:45,500 --> 00:05:50,420 you're getting similar level of performance at half the cost. So yeah, this is a big release for 65 00:05:50,420 --> 00:05:54,280 the SpaceX team because they just proved once again that they are a lab that you seriously 66 00:05:54,280 --> 00:05:59,900 start into considering especially in 2026 and I'm going to talk more about DeepSeek version for Pro 67 00:05:59,900 --> 00:06:05,600 GA but basically what we're seeing today is that both of these releases kind of emphasize the fact 68 00:06:05,600 --> 00:06:10,520 that performance and all above that is the price at what you're getting for that performance is 69 00:06:10,520 --> 00:06:15,880 becoming more and more important for all users because we see many labs now focusing on creating 70 00:06:15,880 --> 00:06:21,380 the best model at the cheapest cost. Last year in 2025, most of the labs were just focused on, 71 00:06:21,440 --> 00:06:26,680 I would say, creating the strongest model. Yes, cost was important, but I think most of the times 72 00:06:26,680 --> 00:06:31,620 the frontier labs, meaning OpenAI Anthropic, were kind of more lenient on that fact because they 73 00:06:31,620 --> 00:06:38,020 didn't have as strong of a competition. Intelligent models that are maybe not always ahead of OpenAI 74 00:06:38,020 --> 00:06:42,560 Anthropic, but match their performance at a fraction of the cost. So yes, price to performance 75 00:06:42,560 --> 00:06:49,800 ratio is becoming a critical I would say indicator in 2026. Now this is the updated benchmark chart 76 00:06:49,800 --> 00:06:54,120 after the release of the new model and one thing you'll see across the board is that it matches 77 00:06:54,120 --> 00:06:58,360 the top level performance of many of the models. The one thing interesting over here is that they 78 00:06:58,360 --> 00:07:02,960 haven't put Opus 5 here for some reason. There is Fable 5 here that we can compare this model 79 00:07:02,960 --> 00:07:08,900 against but one thing you'll notice is that DeepSeek, remember this model costs 43.5 cents 80 00:07:08,900 --> 00:07:14,940 per million input tokens and 87 cents per million output tokens while the other models all over here 81 00:07:14,940 --> 00:07:20,140 are way more expensive than that. So the first thing if you look at Terminal Bench 2.1 the model 82 00:07:20,140 --> 00:07:26,126 scores 87.9. The older version of the model was 72.1 and the flash version was 83 00:07:26,126 --> 00:07:29,526 of the model was 72.1, and the Flash version, which we got last week, was 82.7. And what's 84 00:07:29,526 --> 00:07:37,146 crazy is that Fable 5 is 88. Yes, 88. So this model is only 0.1% behind Fable 5. And then on 85 00:07:37,146 --> 00:07:43,606 the Cyber Gym which is Cybersecurity the model actually beats Fable 5. Fable 5 sits at 83.1 86 00:07:43,606 --> 00:07:49,546 while DeepSeek version 4 Pro the new one that we got today is at 83.3. Now if this tells you 87 00:07:49,546 --> 00:07:54,706 something is that this model is once again really geared at Cybersecurity. We saw the Flash model 88 00:07:54,706 --> 00:07:58,966 also be geared towards Cybersecurity and becoming a model that was quite capable in that area 89 00:07:58,966 --> 00:08:04,986 and we're seeing the same thing today with the new version 4 Pro which sits at 83.3 outperforming 90 00:08:04,986 --> 00:08:10,486 Fable 5 on the benchmark. On deep software engineering this model is not beating Fable 5 91 00:08:10,486 --> 00:08:17,786 because Fable 5 sits at 70 but the model achieves 62.7 which is ahead of GLM 5.2 and a bit behind 92 00:08:17,786 --> 00:08:24,186 Kimi K3 which sits at 67.5 but one thing again this is a big jump compared to the preview version 93 00:08:24,186 --> 00:08:29,686 of the model which was at 12.8 so yeah they have really trained this model and have improved it on 94 00:08:29,686 --> 00:08:34,706 the back end because we can clearly see in this deep software engineering benchmark. And there's 95 00:08:34,706 --> 00:08:40,266 couple of other benchmarks but one key benchmark that this model excels at is the automation bench 96 00:08:40,266 --> 00:08:48,346 where the model achieves 31.8 and fable 5 29.1 so once again the model outperforms fable 5 now this 97 00:08:48,346 --> 00:08:53,166 is a fraction of the cost as i mentioned this is 43 cents and 87 cents for input and output 98 00:08:53,166 --> 00:08:59,606 respectively versus fable 5 which is at 10.50 so yes if we look at the terminal bench for example 99 00:08:59,606 --> 00:09:05,106 we are seeing this model achieve a result which is 0.1 behind the best model out there fable 5 100 00:09:05,106 --> 00:09:12,766 and the pricing is insane to look at 43 cents versus 10 dollars 87 cents versus 50 dollars per 101 00:09:12,766 --> 00:09:19,446 million input tokens so when we turn that into a fraction this model is 57 times cheaper and earlier 102 00:09:19,446 --> 00:09:24,586 last week i think i made a video about how deep seek version 4 pro was going to achieve a result 103 00:09:24,586 --> 00:09:28,866 like this because there was an investor report that leaked and when i saw the pricing where it 104 00:09:28,866 --> 00:09:34,206 said that it was going to match Fable 5 or even outperform it and be at 57 times cheaper per cost. 105 00:09:34,366 --> 00:09:38,906 When I was reading their investor report, I didn't really believe it. But today it is proven because 106 00:09:38,906 --> 00:09:43,786 at least on the benchmark so far, yes, I'm just saying the benchmarks, we are seeing similar level 107 00:09:43,786 --> 00:09:49,086 performance. Now, if this model actually performs like that in production, it's a little too early 108 00:09:49,086 --> 00:09:53,506 to tell yet. We still would have to give a couple of weeks and see how it's performing in the long 109 00:09:53,506 --> 00:09:58,206 run. Because sometimes on the release date, the models perform quite good. But over time, 110 00:09:58,206 --> 00:10:02,406 they kind of deteriorate in their quality. So we hope DeepSeek doesn't do that. Historically, 111 00:10:02,486 --> 00:10:06,726 they haven't done that. But let's just see. Just going to put it out there because right now we're 112 00:10:06,726 --> 00:10:11,886 basing this off of benchmarks. But that's it for today's video. Make sure you guys are subscribed 113 00:10:11,886 --> 00:10:17,666 to the channel. Follow our new newsletter as well at universeofai.beehive.com. As well, 114 00:10:17,726 --> 00:10:22,906 subscribe to the main channel, World of AI, and support us on X by following the Universe of AIZ 115 00:10:22,906 --> 00:10:25,706 as well. Until then, I'll see you guys in the next video.