⚡️GPT 4.1: The New OpenAI Workhorse

We’ll keep this brief because we’re on a tight turnaround: GPT 4.1, previously known as the Quasar and Optimus models, is now live as the natural update for 4o/4o-mini (and the research preview of GPT 4.5). Though it is a general purpose model family, the headline features are: Coding abilities (o1-level SWEBench and SWELancer, but ok Aider) Instruction Following (with a very notable prompting guide) Long Context up to 1m tokens (with new MRCR and Graphwalk benchmarks) Vision (simply o1 level) Cheaper Pricing (cheaper than 4o, greatly improved prompt caching savings) We caught up with returning guest Michelle Pokrass and Josh McGrath to get more detail on each! Full Video Episode Timestamps Part 100:00:00 https://www.youtube.com/watch?v=y__VY7I0dzU Introduction and Guest Welcome00:00:57 https://www.youtube.com/watch?v=y__VY7I0dzU&t=57s GPT 4.1 Launch Overview00:01:54 https://www.youtube.com/watch?v=y__VY7I0dzU&t=114s Developer Feedback and Model Names00:02:53 https://www.youtube.com/watch?v=y__VY7I0dzU&t=173s Model Naming and Starry Themes00:03:49 https://www.youtube.com/watch?v=y__VY7I0dzU&t=229s Confusion Over GPT 4.1 vs 4.500:04:47 https://www.youtube.com/watch?v=y__VY7I0dzU&t=287s Distillation and Model Improvements00:05:45 https://www.youtube.com/watch?v=y__VY7I0dzU&t=345s Omnimodel Architecture and Future Plans00:06:43 https://www.youtube.com/watch?v=y__VY7I0dzU&t=403s Core Capabilities of GPT 4.100:07:40 https://www.youtube.com/watch?v=y__VY7I0dzU&t=460s Training Techniques and Long Context00:08:37 https://www.youtube.com/watch?v=y__VY7I0dzU&t=517s Challenges in Long Context Reasoning00:09:34 https://www.youtube.com/watch?v=y__VY7I0dzU&t=574s Context Utilization in ModelsPart 200:10:31 https://www.youtube.com/watch?v=y__VY7I0dzU&t=631s Graph Walks and Model Evaluation00:11:31 https://www.youtube.com/watch?v=y__VY7I0dzU&t=691s Real Life Applications of Graph Tasks00:12:30 https://www.youtube.com/watch?v=y__VY7I0dzU&t=750s Multi-Hop Reasoning Benchmarks00:13:30 https://www.youtube.com/watch?v=y__VY7I0dzU&t=810s Agentic Workflows and Backtracking00:14:28 https://www.youtube.com/watch?v=y__VY7I0dzU&t=868s Graph Traversals for Agent Planning00:15:24 https://www.youtube.com/watch?v=y__VY7I0dzU&t=924s Context Usage in API and Memory Systems00:16:21 https://www.youtube.com/watch?v=y__VY7I0dzU&t=981s Model Performance in Long Context Tasks00:17:17 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1037s Instruction Following and Real World Data00:18:12 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1092s Challenges in Grading Instructions00:19:09 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1149s Instruction Following Techniques00:20:09 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1209s Prompting Techniques and Model Responses00:21:05 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1265s Agentic Workflows and Model PersistencePart 300:22:01 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1321s Balancing Persistence and User Control00:22:56 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1376s Evaluations on Model Edits and Persistence00:23:55 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1435s XML vs JSON in Prompting00:24:50 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1490s Instruction Placement in Context00:25:49 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1549s Optimizing for Prompt Caching00:26:49 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1609s Chain of Thought and Reasoning Models00:27:46 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1666s Choosing the Right Model for Your Task00:28:46 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1726s Coding Capabilities of GPT 4.100:29:41 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1781s Model Performance in Coding Tasks00:30:39 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1839s Understanding Coding Model Differences00:31:36 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1896s Using Smaller Models for Coding00:32:33 https://www.youtube.com/watch?v=y__VY7I0dzU&t=1953s Future of Coding in OpenAIPart 400:33:28 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2008s Internal Use and Success Stories00:34:26 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2066s Vision and Multi-Modal Capabilities00:35:25 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2125s Screen vs Embodied Vision00:36:22 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2182s Vision Benchmarks and Model Improvements00:37:19 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2239s Model Deprecation and GPU Usage00:38:13 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2293s Fine-Tuning and Preference Steering00:39:12 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2352s Upcoming Reasoning Models00:40:10 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2410s Creative Writing and Model Humor00:41:07 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2467s Feedback and Developer Community00:42:03 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2523s Pricing and Blended Model Costs00:44:02 https://www.youtube.com/watch?v=y__VY7I0dzU&t=2642s Conclusion and Wrap-Up This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.latent.space/subscribe https://www.latent.space/subscribe?utm_medium=podcast&utm_campaign=CTA_2

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:07Alessio Fanelli?

  Hey everyone, welcome to the Lit and Space Podcast. This is Alessio, Partner and CTO at Dasebo, and I'm joined by my co-host swyx, founder of Small AI.

• 0:16Speaker 01

  Hey, and today we have a returning guest as well as a new friend. Welcome, Michelle and Josh.

• 0:22Speaker 02

  Hello.

• 0:23Speaker 01

  Both of you work on the, I guess, Michelle, I think I used to introduce you as manager on the API team. It seems like you've changed your role since we last talked on the podcast.

• 0:33Speaker 02

  Yeah, now I lead a team on the research side, specifically in post-training.

• 0:38Speaker 01

  Yeah, and Josh, you are also on post-training.

• 0:41Speaker 03

  Yep, I'm a researcher on Michelle's team.

• 0:43Speaker 01

  Yeah, and I just found an interesting commonality you guys have. You're also both from Waterloo, continuing the tradition of extremely correct engineers.

• 0:50Speaker 02

  Oh yeah, we talked about that last time, that's right.

• 0:54Speaker 01

  Okay, so we're gathering to talk about GPT 4.1. You launched it, I mean, we got a little preview and it was a little bit rumored, right? It was pre-released, I guess, with Open Router as Quasar Alpha, and then there was also an Optimus version. And I think people were trying to figure out like, why are we going back from 4.5 to 4.1? You know, there's a whole bunch of other things, but like, what are the headline facts? I guess you guys want to emphasize about 4.1?

• 1:20Speaker 02

  Yeah, I'll just say we released three new models today. GPT 4.1, GPT 4.1 Mini, and GPT 4.1 Nano. And the real focus on these, we're just making the models that were great for developers. So we improved instruction following, coding, and shipped our first 1 million context models.

• 1:38Speaker 01

  Josh, anything to add? I don't know if there's anything else that people should really, that are like, sort of in the fine print.

• 1:45Speaker 03

  No, I think the only thing that I would touch on maybe twice is that there's actually a new model in the lineup, Nano, which is even faster for developers that are making, you know, low latency applications.

• 1:57Alessio Fanelli?

  And cheaper. What's the, any fun story behind the code names? Or, you know, I got the strawberry hat as another fun time in the lore of OpenAI.

• 2:07Speaker 02

  Yeah, yeah, we really wanted to get as much developer feedback as possible on this model to make sure it worked well in the real world. And so we tested it kind of through OpenRouter. And it was super cool to see people latch on to the names and, and get the theories going. But the feedback we got from there was super helpful.

• 2:28Speaker 01

  Yeah, yeah, it's not even, it's not even like the name. It's more about just like the API shape. Once we saw like ChatCompo, it was like very obviously OpenAI.

• 2:37Speaker 02

  Yeah, it's a, it's a good note.

• 2:40Speaker 01

  Yeah, but like, I mean, okay, is, is there like an emphasis on stars? Like what, what inference were we supposed to draw from, you know, quote unquote, super massive black holes?

• 2:50Speaker 03

  I don't think there's anything really to draw from there.

• 2:52Speaker 01

  Okay, they're just cool.

• 2:53Speaker 02

  Just fun code names, you know, they make you think of, of cool concepts.

• 2:58Alessio Fanelli?

  The vibes are good. The vibes are good.

• 3:00Speaker 02

  You know, yeah.

• 3:01Speaker 03

  Yeah.

• 3:02Speaker 01

  The other thing about the examples, we're just mining for lore here, right? The interesting animal comes up a few times on the live stream and on the blog posts. What's up to Tapir's? Who, who likes Tapir's here?

• 3:11Speaker 02

  Yeah, our, our team is just a super big fan of Tapir's. So

• 3:15Speaker 01

  Okay.

• 3:16Speaker 02

  They just happen to work their way into a lot of our content.

• 3:19Speaker 01

  Okay, cool.

• 3:20Speaker 03

  Awesome.

• 3:21Alessio Fanelli?

  Yeah, go ahead. I, I, I think like the first thing that, yeah, we just want to run through is obviously the 4.1 to 4.5. I, I think that's the first thing that everybody was maybe confused about. So, and I know you're debrogating 4.5. It sounds like 4.1 is just like a kickass model. And the 4.5 size, maybe it's not as good of a fit. That was just a research preview. So, yeah, I, I don't know. Whatever you want to say to, to address that, I think it's something we've seen come up also in the Discord.

• 3:50Speaker 02

  Yeah, totally. Okay, naming is really hard and we've tried to make this as less confusing as we can. But you know, nothing's perfect. Basically, the way we got here is that GPT four point one is like a pretty big improvement over the four O line. And we really wanted to signify that. However, it's a model that's like much smaller and cheaper than GPT four point five. And as a result, you know, doesn't achieve the same like Amy or other intelligence evals. So it doesn't beat four point five on all of the evals. And so we didn't think it made sense to increment beyond four point five. But we do think for most developers, they can kind of replace a lot of their four point five usage with four point one.

• 4:32Alessio Fanelli?

  And then the mini is strictly better than four O mini.

• 4:36Speaker 02

  Yeah.

• 4:36Speaker 03

  Yeah.

• 4:39Speaker 01

  With the with the nano like, but like, we don't know if four point one is a dissolution of four point five or there's there's no relationship there. Like, what what can we say about like the shared linear?

• 4:50Speaker 02

  Yeah, what I'll say there is we're always using various research techniques to improve our models. And distillation is something we talked about before. It's it's really meaningful, especially for the small models. And we've kind of pulled out some of the things that made four point five really good. Like it has a lot of the instruction following greatness and also rolled that into four point one.

• 5:12Speaker 01

  Awesome. I think one of the, because I strongly remember on the four O launch that their communication was that we're kind of moving to a new model architecture that is Omni model. Right? That was what that's the O in four O. And then four point one is part of this subsequent trend of trying to merge everything, like the reasoning model, the Omni model, everything. And I think that there's there's this doubt about whether four point one, I I think it's basically trying to be sold as a strict replacement for four O. But I don't know, i- is it going to be fully Omni model? Is it? Is it like roughly the same architecture that we that we that we think four O has?

• 5:51Speaker 02

  So we we already have different slugs on the real time API and like responses API. So they're already, you know, somewhat different checkpoints. I don't think we don't have any current plans to release four point one in the real time API. But you know, things things may change.

• 6:07Speaker 01

  Yeah, and then there's ImageGen and all that, right? Like, so, and as far as we know, no plans, maybe, but nothing announced.

• 6:14Speaker 02

  Not, not right now. The focus for 4.1 was kind of these three core capabilities for developers.

• 6:20Speaker 01

  Yeah. We, our Discord actually also did a launch watch party for the recent 4.5 podcast that Sam Altman did, where I think for the first time, it was basically kind of confirmed that like some, something that people already knew, like Andre Karpathy was already talking about this, that 4.5 was like 10x the size of four. And I think there's a question about like, do we do the linear interpolation of 4.1 is like, you know, zero point, is like, I don't know, 2x the size or something?

• 6:48Speaker 03

  That's not really how we think about naming the models. There's a whole bunch of parts that go into the, the recipe. And so, you know, it doesn't really reflect on just the pre-training recipe, our version numbering. But I think the 4.1 is just because of the large jump that we have in like coding capabilities, long contexts and so on. It's more so what it's like for the end user, more so than anything about the training recipe.

• 7:09Speaker 02

  We, we can go a little under the hood on training, though. And we'll say that, um, you know, Nano is obviously a new pre-trained. We also have a new pre-trained for Mini. And then the larger version is a new mid-train. But we find that actually a significant amount of the gains come from new post-training techniques. And so I think in the past, the narrative is that you need to pre-train these larger and larger models to get better performance. And we're finding that we're able to squeeze a lot more out of post-training now.

• 7:40Alessio Fanelli?

  Talking about how big a model is, the other side of it is the context window. You have a 1 million context. I know that Sam at that day last year, he said that, yeah, 1 million was like months away. So right on, right on time. Can you talk about, yeah, how hard that was to get to 1 million? And then maybe where the end game is in your mind? Is it 10 million, 100 million, infinite? What, what, what really matters as you start to scale this?

• 8:09Speaker 02

  Yeah, Josh worked a lot on long context, so he's the right person to ask.

• 8:12Speaker 03

  Definitely. So I think the first thing that we that I thought was really interesting when we were going to long context is actually some of the evals that you see as like headlines on maybe other blogs where it's needle in a haystack. Actually, most of the models do really well right out of the box, but then we had to actually first get a lot of measurement on the longer context for long context reasoning. So you know, we actually just open sourced two new evaluations that are about using the context in a more complex way. So you know, one of them, you have to reason a lot about ordering and the other is actually walking through graphs. So there's a lot of reasoning that you have to do in those data sets and that's where doing long context is actually much harder, but single needle in a haystack, we were able to

• 8:53Speaker 03

  saturate pretty easily. And then the most of the work came at this these harder tasks.

• 9:00Alessio Fanelli?

  Yeah, I was gonna say, how not just how you think about length of context in terms of consuming documents versus like active, kind of like thinking and planning. I think there's obviously a whole part on the prompting side around billing a Agentic workflows. Do you do you think that people maybe like still think too much of it about, yeah, needle in a haystack kind of document retrieval versus like traversing very long plans and kind of like iterations in context?

• 9:26Speaker 03

  Yeah, I think the mental model that I have is maybe actually has some more variables in it. So there's single, there's the needle in a haystack where you have like some amount of distractors and some, you know, needles that you're trying to find. And I think that it's more so about how dense of the context do you need to use? So like summarization, you're you're actually just using the entirety of the context, whereas you know, needle in a haystack, it's very sparse. And then I also generally think about orderedness. If you're going to make some sort of inference on this, do you are you just look looking, you know, sort of front to back? Or do you need to move around in the context in order to generate a good answer while the model's sampling?

• 10:06Speaker 01

  Yeah, is that something that you worked on with Graphwalks? Is that the?

• 10:09Speaker 03

  Yeah, that was sort of the the most synthetic and clean way to measure the model. And then, you know, we worked on a lot of other training techniques, data to sort of test the model's ability and train it in the model's ability to reason throughout the context in a sort of shuffled way.

• 10:25Speaker 01

  Yeah, you know, actually, I have the ability. I like to give people a little bit of visual aid with with these things. So I actually went into your Hugging Face release and got an example of the graph task. And so there's there's a few versions of this, right? There's like the BFS and DFS version. And also, it's I guess it's very character specific. So I don't know, maybe could you tell us like, you know, design choices around this? Like what was surprisingly hard and anything like that?

• 10:51Speaker 03

  Yeah, so the idea here is you you take a graph and you encode it into the context by looking at the the edge lists and just putting that into the context and then asking the model to do an operation. And then, you know, under the hood, we're actually just executing the real operation and using that to then evaluate the model's ability to work. One of the things that I found surprising at first was the what the model would do when it wasn't sure how to use its context, you know, early versions of the model just sort of looping, saying like, oh, no, I can't find this edge that I think should be there. And, yeah, I think I was actually very surprised how all models seem to have more difficulty than I would have expected on a task that, you know, we would find very simple or like,

• 11:35Speaker 03

  you know, maybe an undergrad could write a Python script to to run in a couple of minutes.

• 11:42Speaker 01

  Yeah, right. Okay, so like, what is the like, what is the real life task that this is meant to model? I guess, you know, I feel like the other one, MRCR.

• 11:54Speaker 03

  Yeah.

• 11:54Speaker 01

  Seems a little bit more intuitive where, you know, you have like four different stories and you pick out the second one. And that's a that's a real task that the people have, but people don't really traverse graphs. Like, this is a bit more theoretical, but like, you know, is there was there any sort of correlation study done?

• 12:09Speaker 03

  Yeah, this is actually meant to be sort of the idealized version of like a multi hop reasoning benchmark. So we have a lot of things where you know, you're putting hundreds of documents into the context and then you might ask a question that you actually have to traverse 10 documents for. But there the edges are, they're implicit, right? Like there is some underlying graph that's connecting all of these documents that you need to traverse in order to answer the question. But they're actually much harder to traverse because the edge isn't actually given to you. And so the question there was like, okay, if I actually just give you all of the IDs of these things that you need to traverse, can the model even do that? Where it's like, it's actually just a lower bound on how well the model can do. And I think it's, that's actually somewhat well,

• 12:52Speaker 03

  well reflected in some of the internal benchmarks we have that are using more natural data.

• 12:56Speaker 02

  Imagine something like a tax return, right? Where you like upload the entire tax code. Like to figure out what to put into this, you know, box, you'll need to reference all of these boxes. And so this is like a similar level of multi hop reasoning. But again, like Josh said, all of the references are implicit.

• 13:15Speaker 01

  Yeah, I think that like some kind of backtracking, if it's needed, is also super interesting, especially for agent work. I, for listeners who've been listening to us for a while, we actually covered this paper at New Earth's last year, two years ago, called Gival, where they actually modeled graphs for graph traversals for agent planning. And it reminds me closely of that. It's just that they never came up with this exact format that you have here, which basically is the same thing. I also like that you included blank answers because sometimes people do hallucinate, or models do hallucinate answers and you have a fair amount of blank ones.

• 13:50Speaker 03

  Thanks for random sampling over graphs, I did, I guess.

• 13:55Alessio Fanelli?

  Yeah. Is this tied also to the file search API that you released recently? Like, how should people think about how everything kind of comes together in the API?

• 14:06Speaker 02

  Yeah, I think oftentimes with retrieval, you, you know, might be using rag to fill the context. And a lot of this is like to get around the limitation of a short context window. So we do expect a lot of developers to start, you know, uploading their full context more directly to the model. So for smaller tasks, you maybe don't need the whole vector store. Um, but we do anticipate this to play well with that paradigm as well. Like, maybe you can just insert way more chunks into the context. So we, we think it'll play nice.

• 14:38Speaker 01

  Yeah. A- any relationship to the memory upgrades in ChatGPT that we recently got? Is long context just directly usable for memory? Or should we just always have a separate memory system?

• 14:51Speaker 02

  Yeah, it's a, it's a good question. So right now, the dreaming feature, we kind of have some of these memories embedded in the context. But, you know, they are, they are separate features. So four point one is powering the API, whereas the enhanced memory is, is ChatGPT only.

• 15:11Speaker 01

  Yeah. Awesome. Yeah, I think, I think that's interesting. I guess the, the one last thing I'll call out on long context, which is kind of unintuitive, or maybe there's an explanation, which was the, you, you had two needle for MRCR. And then we had four and eight. And everything kind of just regresses to some kind of baseline of like, let's say thirty percent or twenty percent as that. But it's interesting to see where the smaller models sometimes match or outperform the larger models. Uh, I was wondering if there's anything unusual there? Or do you think it was like a bad roll of the dice?

• 15:45Speaker 03

  I think it's probably just a, a bad roll of the dice. I think I would probably look more so at the, the larger.

• 15:49Speaker 01

  The narrow ones.

• 15:50Speaker 03

  Few things regress as you increase the number of needles, because there's sort of more complex reasoning that it has to do about the order of different things in its context.

• 15:59Speaker 01

  Awesome. Yeah, cool. Happy to, happy to move on from there. Yeah, we have a whole bunch of other evals that we can go over. So I had in my notes that we could talk over, you know, anything that you, that you want. There was also like Kali from Shunyu with, who we have, have on a podcast for instruction following. And I, I realized that, you know, he joined OpenAI and I wonder if he had a role to play in that one.

• 16:22Speaker 02

  No, we did not collab, not on it. Honestly, I think it's best when eval authors and model developers don't collab too much. Cause you, you want things, you know, as objective as possible, not trying to gain any evals.

• 16:35Speaker 01

  Yeah. And then I think there was also like, for the first time, the announcement of the, or shout out of the internal instruction following benchmark from API data. People have had the ability to opt in to share data for a while. Actually, I like published the, I posted a tweet, you know, cause I found it in the dashboard that you can just opt in and, and like, there's, there's basically 16 days left for this program where you can just get free inference. And like, so I'm just kind of curious, like what you found from that kind of IF eval that, that might be different from the normal IF eval that people have.

• 17:08Speaker 02

  Yeah, totally. A lot of the instruction following evals that are open sourced are open sourced, you know, or crafted in a way that are easy to craft. So for example, like Graphbox is, is somewhat easy to craft. Like you can create this graph and verify it easily, but it is not exactly aligned with what the users are doing. And this is true for some of the instruction following evals where you ask the model to output exactly four words or, you know, three paragraphs or stuff like that, things that you can verify easily in code. And these are useful instructions, but we find that many of the really interesting instructions are actually challenging to grade. And so the open source evals often don't have them.

• 17:50Speaker 02

  And so getting this like real world diverse set of data actually helps us find like, what are the commonalities in what developers are doing? What is a really good example of like a negative instruction? And then we can go from there and figure out how to, how to evaluate it.

• 18:07Alessio Fanelli?

  Yeah, I think that's, there's also an interesting question of like, what domains do people use you on? And I wonder if like, there's a way to tell you.

• 18:17Alessio Fanelli?

  Because sometimes it can be very confusing if I, especially because maybe I'm building an app and letting people use my key, but other people are building apps on top of me. So you have just a lot of chaos of like, multiple degrees of abstraction, where you just have to parse through the prompts.

• 18:36Speaker 02

  Yeah, it's true. Well, I will say, we do use our own products internally where we can. And so we're not manually by hand reading every prompt. After they're like anonymized, we scrub them of any identifying data. Then we use our models to take passes to categorize them. And so if we get feedback that like, we're not doing well on ordered instructions, then we can kind of do a pass over all of our data and find some good examples of those.

• 19:02Alessio Fanelli?

  So there's the instruction following section in this great prompting GPT-4.1 models. I think maybe we can go through some of these examples. The first one that caught my mind that it's not necessary to use all caps and other incentives like bribes or tips. But developers can experiment with this for extra emphasis. So I think the second part leaves me confused. Are you saying that people should still try and do this? And sometimes the model responds positively to it? Do you feel like it's still just part of the lore? I'm curious why. I would have loved for you to say either yes, it works. Or like, no, you should stop. It looks silly. I guess the truth is somewhere in the middle.

• 19:40Speaker 02

  The truth is always messy. Reality is that our models have gotten a lot better at following instructions just stated once and clearly. But we find, honestly, developers often become the best experts at prompting our models. Because you know, you're building your livelihood on this thing and get to know the details of it really intimately. So I will say, stuff like that won't hurt the performance of the model. But we kind of always want to leave it open to people to figure out what works best.

• 20:08Alessio Fanelli?

  Yep. Yeah, and then you had to always start with a response rules or instructions section. Are those keywords meant to be taken kind of like verbatim? Like those are kind of like the tokens that work the best? Or is it just like a, an example?

• 20:23Speaker 02

  Warm example. Yeah.

• 20:25Alessio Fanelli?

  Okay, cool. Yeah, this is great. I feel like until today, we did an episode with like the prompt report on like all these prompting techniques. But then it's also unclear for which model, which ones work best. So it's super useful. And then you had a, in the agentic workflows one, you have a persistence thing. It's like, please keep going. How much? And I, and I think I read that improves like the sweet, the sweet bench like 20%? Just by having like the persistence.

• 20:53Speaker 02

  I wouldn't, it's not that this one prompt improves we bench 20%. It's that we found this is the most effective harness for our model. And combined with all the post-training improvements, it results in the big improvement. But yeah, like the model is trying a lot to be helpful and often it wants to check back in with the user. And like, you know, should I keep doing this? Like, am I on the right track? And so a prompt like this makes sure it keeps going, doesn't bother you again, and just gets the task done.

• 21:22Alessio Fanelli?

  Yep. Yeah, I think like there's this interesting trade off between persistence and yielding back to the user. The more agentic a model wants to be, the more persistent it should be. But then sometimes it just goes off the rails. And I wonder how you solve this trade off? Because sometimes it just goes too far. There's been criticisms of Claude Sonnet trying to rewrite too many files at once when I just wanted to make one thing. For example, and that's a, that's a form of bad persistence. I, what are the axes here in which like you think about it?

• 21:57Speaker 02

  Yeah, I think one interesting thing that comes to mind here is that we had an extraneous edits eval. Where you asked the model to make an edit and classify like were all of its changes related to what it was asked to do? Or did it go off and, and do a little too much? And we found that from 4.0, which got 9%, it's pretty crazy. 9% of the time making an extraneous edit is a lot. 4.1 is at 2%, so it's a pretty big improvement. So yeah, I'll, I'll just say like focusing on this, we've heard feedback about this. We made an eval and we made sure to track it and improve it during training too.

• 22:34Speaker 01

  Yeah, yeah, I mean, everything comes out the evals as as as no surprise to anybody.

• 22:39Speaker 02

  That's true.

• 22:40Speaker 01

  There's another interesting eval that I think is causing some noise. For the first time, I think also that and you being the master of structured outputs should know that JSON is bad now and we should all use XML.

• 22:53Speaker 02

  I wouldn't say that. I don't know which eval you're talking about, but.

• 22:56Speaker 01

  It's in the prompts guide, which maybe you guys didn't write, so we kind of spring this on you.

• 23:01Speaker 02

  Yeah, Noah and Julian and our team wrote the prompt guide and did a great job. I do think XML is very helpful for structuring prompts, whereas for parsing outputs, maybe the story is a bit different. Like, sometimes it's really useful to get outputs in JSON, so you can plug them directly into your application. But I do think the models work particularly well with XML as inputs. But Chris, you need that?

• 23:25Speaker 01

  No.

• 23:28Speaker 01

  Cool. I mean, I think people always just care a lot about code tool calls and structured outputs, as you well know. And so any updates to instructions over there is good. People also are interested in this concept of that apparently putting the instructions and user query at the top and the bottom, so duplicating it at the top and the bottom in the context is much better than is better than putting it top only and much better than putting it bottom only. Again, this is from the prompt guide, so I don't know how aware you guys are on this.

• 23:58Speaker 03

  Yeah, I think part of that was just like, you know, here at Go, we we tried all three for when we were evaluating the model and having that redundancy is definitely the best. But then using those the instructions at the beginning, the model is going to be able to then take that into account as it does processing.

• 24:14Speaker 01

  Yeah, I think like a lot of people would see this as like running counter to prompt caching, because obviously you want to put the things that change a lot at the bottom. Basically, is this fixable in post-training? Like, can we just tell models to take instructions or user queries only at the bottom? Because we want to optimize for prompt caching.

• 24:38Speaker 02

  When we figure it out, we will do that.

• 24:42Speaker 01

  I mean, it seems doable. It seems like a post-training thing. I don't know. Maybe my mental model post-training is wrong.

• 24:47Speaker 03

  So I think actually having things at the the beginning of the prompt, you would still get prompt caching there. If you're putting in, for example, like a big needle on haystack and you have the data changing each time, like either per user, there's still different ways that you can be putting the prompt at the beginning and getting a lot of the the cache hits. It sort of just depends on your use case.

• 25:06Speaker 01

  Yeah, awesome.

• 25:08Alessio Fanelli?

  The the other thing I noticed, I know you made a note of this, Sean, too, is our chain of thought and reasoning and how people should think about this model versus a reasoning model. Yeah, what's your, I, yeah, should I just use 4.1 and prompt it to do a chain of thought? Should I use O1 and make a plan and then use 4.1 to implement the plan? How should people think about composability?

• 25:32Speaker 02

  Yeah, it's a great question. We have found that 4.1 is a lot better at doing planning and thinking through its steps in COT when prompted than our previous non-reasoning models. But our reasoning models are designed to have kind of more coherent plans, be able to reason over longer horizons than these non-reasoning models. And you can see that reflected in things like intelligence benchmarks. So Amy, GPQA, stuff like that, you'll see the reasoning models do much better. So in general, I would say, like, the question you're really getting at is like, I'm a developer, which model should I be using? And I think the answer is always going to be the fastest model that accomplishes your task, right?

• 26:16Speaker 02

  So maybe you start prompting 4.1 as a starting point. If it does your task super well, then maybe you drop down a 4.1 mini and save latency or even nano. Whereas if 4.1 is struggling a bit little, maybe needs more coherent reasoning over longer time horizons, then maybe you upgrade to a reasoning model.

• 26:35Alessio Fanelli?

  Is there a quick way to get through this here at six? I know one thing that a lot of people do is like they use O one for like a plan and then they put that plan in cursor and then have the plan applied to their code base. It sounds like there's maybe not a rule to when to do which it just like task dependent.

• 26:54Speaker 02

  Yeah, I would say we're all kind of figuring out the best way to use these models together. And so I do think reasoning models for planning and using kind of more targeted models to execute is definitely a good architecture.

• 27:06Speaker 01

  Cool. If there's nothing else on on that side, I'd love to go into the coding, which is something that we're emphasizing a lot. It's doing super well. It's better than O one and Sweet Bench. Was that expected?

• 27:17Speaker 02

  Uh, not really.

• 27:20Speaker 01

  Yeah, like what's the, I mean, what's the story there? There's also Sweet Lancer, which is a newer one, which attaches a money value to things. And basically, what like what should people understand is going on here? Like, is it a better coding base model or just a coding agent model? And I think there's also, there's also a question about like, you know, how important to coding is it if I'm not using a coding use case?

• 27:45Speaker 02

  Yeah, so I'll start by saying we just set out to make a model that was great at coding, both in your terminal or in your editor or wherever you want to use it. And so we kind of broke that down into the problems that it, you know, encompasses. So like developers want the model to produce better diffs, for example, or they want the model to explore the code base correctly, or they want to produce code that compiles or produce code that writes tests. And so our approach was kind of teaching the model all of these various facets. There's kind of just a bunch of work streams that all coalesced around GPT four point one.

• 28:21Alessio Fanelli?

  Yeah, I think much improved post-training all over to make for a better coding model.

• 28:28Speaker 01

  Yeah, I think there's like different kinds of coding, right? Like it's interesting for me to observe that there, for example, so I'm just gonna pull it up on the chart here because I always like to show people visuals. You're fifty five on Sweet Bench and O1 gets like a forty one. But then on, oh, I don't think I, I don't think I have the others. Like, but, but ADR is, it is less, it is not at O1 level. And so I, I think, I think I struggle to get some kind of intuition of when, like, what are the different elements of coding? I guess there is like, you know, single file edits, whether it's like a diff or a whole file. And then there is entire project edits. Is that a reasonable split? Are there more to this?

• 29:11Speaker 02

  Yeah, that's one way to think about it. Basically, where GPT four point one, can it kind of explore, go through a repo? It's been trained to do that particularly well. Whereas, you know, to just get some code and produce a change, a reasoning model might do better because it can kind of reason over the entire file. And so that's one good way to think about it.

• 29:34Speaker 01

  Yeah, yeah, that's fair. Any, any understanding of like the smaller ones, the smaller models? Like basically for coding, I should only use four point one and forget the rest?

• 29:45Speaker 02

  You, you might like want to use the smaller models. Maybe if you have like, if you have an IDE where you need an autocomplete feature, for example, or if you want something super fast, if you're building like, I don't know, a text to SQL thing, you might want the first version to populate instantly. So you can see like four point one mini is actually quite significantly better than four O mini, but not that far away from the old four O. So I do think that model will find use case in a bunch of these coding niches.

• 30:17Alessio Fanelli?

  And I know you might not be able to talk about this, but the clip of the, of an AI CFO talking about the Agentic suite has been going viral, I think, today. It, it seems like every lab is putting a lot of emphasis into coding. So, yeah, I'm just curious if there's anything you can share about how people should think about OpenAI in coding? You know, obviously today you don't have. You know, Claude, Claude, you don't have any, anything related to coding. And I think the WinSAR partnership today, they're giving 4.1 for free, 4.1 for free. For a couple of weeks is maybe like one of the first OpenAI endorsement, I guess, on the live stream. But yeah, just, I know there might not be an answer that the PR team might approve. But I'm curious if you have any takes and thoughts.

• 31:00Speaker 03

  I think just stay tuned.

• 31:02Speaker 02

  Yeah, we, I think coding is an important use case for our users. And so that's why we focused it on a lot for 4.1. We also love to use our own products internally. And so making 4.1 selfishly helps us move faster as a company. And so that's where the real focus has been for this model.

• 31:22Alessio Fanelli?

  Do you track what percentage of code is written by 4.1 internally now?

• 31:27Speaker 02

  We do have some metrics like that. I don't have it off the top, but I was actually just talking to one of the researchers on the team who worked on something over the weekend. And he said that this model, GPT 4.1, was able to like get 4950 of his commits on this massive PR. So we were pretty happy to hear that.

• 31:46Alessio Fanelli?

  Excited to use it. Awesome. Yeah, I think on the, yeah, I, I think coding is a super exciting use case. And I think like OpenAI has always been very developer first, as you've been to Michelle. So it's good to see the, the convergence.

• 32:00Speaker 01

  Yep. The other, I think the last capability that I kind of vectored in on was vision or just multimodality in general. It is a lot better. Basically, I, I, I think like, I, I really like these niche benchmarks like MathVista and ChartScive. Yeah, just any, any, any extra color on, on like the vision side that you wanted to talk about, but maybe you couldn't fit into the blog post.

• 32:23Speaker 03

  Yeah.

• 32:24Speaker 02

  Yeah, go ahead.

• 32:25Speaker 03

  Oh, I was, I think one maybe small nugget there is actually, I think the, that 4.1 mini is really exciting on that front. As we were talking about, it's a different pre-training phase. And I think that really shows up in some of the vision evals.

• 32:39Speaker 02

  And yeah, we talked about like coding instruction following along context, a lot of gains coming from post-training. But in particular multimodal, like basically everything you're seeing, the gains are there from pre-training. So kudos to the pre-training team there. They've done incredible work on, on perception and multimodal.

• 32:57Speaker 01

  Yeah, totally. Something that we've been exploring on the podcast for a while, and I'm curious if there's any takes on your side, is is there a strong split between like sort of what I call like screen vision versus embodied vision, right? Like, are you taking pictures of, are you training on snapshots of a computer for computer use, or, you know, and anything with charts, anything like on a PDF is very similar to that, or pictures from the real world, which is more embodied, right? Like where a robot might be able to use that. People have argued back and forth. I'm curious where the movement is or the emphasis is.

• 33:36Speaker 03

  I think one of the, first off, I think that 4.1 is better at both of those things, regardless of how it was actually trained. I think I would probably somewhat defer to the pre-training team when it comes to which one you should be using. We're using, you know, a mixture of both, but we've improved our results across evals on both.

• 33:57Speaker 01

  Awesome. Yeah, that's something that I think people should, definitely do want to explore, the more embodied stuff as well, because the benchmarks tend to focus on the the screen vision stuff, you know, more more chat, more controllable.

• 34:08Speaker 02

  It's always easy to evaluate that is easy to grade.

• 34:10Speaker 01

  Yeah, exactly.

• 34:11Speaker 02

  Those are the things that get looked at the most.

• 34:13Speaker 03

  Yeah, I think one of the things that was really funny with both of the 4.1 mini and nano is we had some strange internal eval results, and it turns out that actually the, these new vision capabilities, they were able to read like, you know, signs in the background and stuff, which was actually changing like some of the validity of our results. And so we were, you know, just running into different eval problems as you actually improve the models.

• 34:38Speaker 01

  Is there a feature of a 4.1 image gen? Or is that like a completely different part of this vision? Like, you know, in some sense, vision is image to text, and the other the other way around is image gen. Is it that simple? Or is it something else?

• 34:52Speaker 02

  It is not. No place right now to to get 4.1 image gen.

• 34:57Speaker 01

  Well, you know, it's very, very popular.

• 34:59Speaker 02

  We like it.

• 35:01Speaker 01

  It's like melting your GPUs. I mean, talking about GPUs, right? Like, you know, part of this whole deprecation of 4.5 and moving people to 4.1 is to get back your GPUs. That's a that's a message that both Shiki and Kevin Wild have mentioned. But like, you are running all these models concurrently for the next three months. Like, I don't know if you get back a GPU, so you think you just grow the usage even more?

• 35:23Speaker 02

  Yeah, I do think, you know, people get the message on deprecation and start moving over. And so what as developers use this model less, we can kind of reclaim that compute. But you're right, it takes a while. And the trade off there is really our commitment to developers. Like if we have something in the API, we won't take it away on without, you know, sufficient notice.

• 35:44Speaker 01

  Yeah, with some notice.

• 35:45Speaker 02

  That's the trade off that is right for us.

• 35:48Speaker 01

  Okay, awesome. Then a couple other smaller announcements. Fine tuning available day one, which is, I think, new for OpenAI. Usually you have to wait like a month or two for the fine tuning capability. For one, four point one only and mini, four point one and mini only and nano and future. Any specific callouts for for fine tuning? I guess like this fine tuning is general discipline that always applies. But any wins that you guys can talk about?

• 36:14Speaker 02

  So first off, yeah, shout out to the fine tuning team. They've worked really hard to get this ready on day one. One thing I will say is that I think people have slept on the preference fine tuning offering. Or the, I think that's what we call the product.

• 36:28Speaker 01

  Yeah.

• 36:28Speaker 02

  So SFT is, people know it pretty well. It's the original fine tuning we had. Whereas this preference fine tuning is super helpful for steering in a particular style. And so I think not enough people are using that.

• 36:41Speaker 01

  Isn't that only for reasoning models or is that for everything?

• 36:44Speaker 02

  No, that's reinforcement fine tuning is only for reasoning models.

• 36:47Speaker 01

  Right.

• 36:47Speaker 02

  Preference fine tuning will offer the pairs. Yeah, exactly.

• 36:51Speaker 01

  Yeah. And I thought it was in alpha. This is why I haven't looked into it.

• 36:55Speaker 02

  I thought it was.

• 36:56Speaker 03

  I think it's RFT that's still in alpha.

• 36:58Speaker 01

  Okay. Well, that's a lot of confusion that we that we just cleared up. Yeah, I think we're going to, you know, I'm doing my conference again in June. And I think we're going to do a workshop on just general, all the fine tuning options. And I think that will clear up a lot of things, which is good. Okay. New models. I know that we can't talk a lot about a lot of them. Noam Brown from your reasoning team just said that there should be a follow up on reasoning models soon. What can we say about it? Sounds like he's.

• 37:25Speaker 02

  We're not the right people to ask, but but stay tuned for.

• 37:29Speaker 01

  Yeah, but like four point one is a good basis for like whatever comes next, right? Like.

• 37:33Speaker 02

  Yeah, not all of our models kind of build on each other necessarily, but we think 4.1 is a great standalone offering for developers. And we also think reasoning models are a good tool and toolbox.

• 37:46Speaker 01

  Yeah, like more just generally, like I always want to explore the relationship between non reasoners and reasoners. And then also like how we merge them, are we doing routing, you know, anything, anything of that sort. Obviously, you have a lot of secret sauce. Cool. And then I think the other thing that a lot of people are demanding or asking about is the creative writing model. Will that ever see the light of day?

• 38:07Speaker 02

  We're working on incorporating kind of those improvements into the models more generally.

• 38:12Speaker 01

  Not as separate, at least.

• 38:14Speaker 02

  People loved about 4.5 is like the humor, the green text, the nuance. So we've heard that feedback and I know, yeah, there's lots of folks working on that and trying to bring it into our next models.

• 38:27Speaker 01

  Awesome. Alessio, anything else?

• 38:29Alessio Fanelli?

  No, this was great. Any requests for the developer community? Things that you want them to try out that maybe people are not doing, things you want them to build for you using the new one, the new APIs?

• 38:40Speaker 03

  I feel like first off, send us feedback. It was really useful to look at different partners and customers who are using our models and to get this like nice wrapped feedback from them, it allows us to iterate a lot faster.

• 38:54Speaker 02

  And on that vein, you know, opt in to data sharing. This just helps us make the model better for you. And one kind of slept on way to do this is the evals product. So you can upload an eval and opt in such that we'll pay for the inference costs if we can also use the eval. And this is just another great way, like we'll, we'll use those evals to make sure our models are, are getting better for people over time.

• 39:19Speaker 01

  Yeah, I think the evals is permanent, there's no end date announced, but the opt in in the API is at least until April 30th. I think a lot of people still don't know about it. We might want to extend that so that people can do more of that.

• 39:33Speaker 02

  Good flag, all, all raised with the team.

• 39:35Speaker 01

  Yeah, awesome. And I think the last question I had was on just on pricing. I think pricing, you know, it's basically just generally cheaper than 4.0, but like not a ton, but like cheaper. And then you're also introducing this concept of blended pricing for the first time that I've seen it. But maybe it's just been out there for a while because you have caching and all that. Just generally, what is the cache to non-cach ratio that we should be thinking about when thinking about workloads? Like, is there a general rule of thumb?

• 40:06Speaker 02

  So one clarification, which is that GPT 4.1 mini is not cheaper than GPT 4.0. So not, it's not just like a blanket decrease in all the models. But however, 4.1 mini is cheaper than 4.1. Also, not sure if this is widely reported, but we've increased our prompt caching discount from 50% to 75% on these models.

• 40:28Speaker 01

  Yeah, I saw that.

• 40:30Speaker 02

  So that's a big input, you know, into figuring out what kind of application you build. And then your question was on, like, what kind of

• 40:38Speaker 01

  Yeah, blended pricing, right? Like, I think there's this question of comparability of prices across models and across providers. Because like, I, you know, it's like, some people are three to one in terms of context to output. And then some part of that is cached. I selfishly, I make a chart that just plots all the model labs versus all the prices. And I'm sure you guys have seen it. And I don't know what numbers to plug in there. So what are people seeing in real life? What's the median, you know, caching rate?

• 41:10Speaker 02

  I don't think we have that off the top. The blended pricing is more to just make it easier to compare. Like, so you could say something like GPT 4.1 is 25% cheaper than GPT 4.

• 41:21Speaker 01

  Yeah, you want one number.

• 41:22Speaker 02

  Yeah. Yeah. Yeah. No.

• 41:26Speaker 01

  All right, well, I'll have to figure it out. But thank you so much. That was fantastic. Thanks for all the work. I think people are very excited to get to work testing this out, giving you feedback. And I'm sure we'll be back again for the next one, probably the reasoner.

• 41:39Speaker 02

  Nice. Thank you, guys.

• 41:42Speaker 02

  Thank you.