OpenAI are testing a new model on the Web Dev Arena Arena under the name 'Anonymous Chatbot 0717'. I can't believe I'm gonna say this, but it is genuinely at a completely different level of front end coding - far better than Sonnet, o3, Gemini 2.5 Pro, or Grok... 4. To test it, I ran a great prompt borrowed from the amazing The Feature Crew YouTube channel, asking models to create a procedurally generated planet with Three.js. Take a look yourselves, but I'm pretty astonished by how big the jump is. I have featured the new model twice just because its implementations have been so interesting. Of course, this is only one test, but OpenAI models have always been a bit 'meh' at front-end work, and they seem to have finally overtaken everyone else on that front. We'll see when it comes out. Credit to Chetaslua for discovering the modelshow more

@lmarena_ai It’s probably the o4 model that’s the full sized version of o4-mini. Apparently they’ve had it for a while and just been doing safety testing and other stuff with it.

@lmarena_ai It is tagged as o3-alpha on the arena

CC: @btibor91 @scaling01 @altryne @koltregaskes @kimmonismus @chrisgpt @apples_jimmy @testingcatalog Anonymous Chatbot 0717 side by side

You should subscribe to @FeatureCrewPod YouTube Channel, they are doing great work testing models on a bunch of different coding & analytical tasks

@lmarena_ai I also made a open source project using Claude 3.5 last year and it looks like this, the repo on GitHub as well:

@lmarena_ai Great job 👏 and yeah it's o3 ALPHA

@lmarena_ai I don't see this model on LM Arena?

@lmarena_ai It would be in the Web Dev Arena and it would be in the battle mode, so you need to get lucky

@AIandDesign @lmarena_ai So it's 1 out of 40 or more odds that you will get this model?

@AIandDesign @lmarena_ai No they prioritise new models, it is more like 1/4 chance I'd say

@lmarena_ai how do we know it's o3 and not o4-full?

@lmarena_ai It is tagged as o3-alpha on the arena

@lmarena_ai that could just be the test name though. o4 would be a continuation of o3, so maybe early checkpoint?

@lmarena_ai the song is a banger

@lmarena_ai Full version:

@lmarena_ai thanks

@lmarena_ai Awesome post! Horrible music tho

@lmarena_ai Haha, I kind of like it though, gpt-4.5 + Suno

@lmarena_ai what was prompt?

@lmarena_ai here it is

@lmarena_ai nice

@lmarena_ai Lol I saw the same earth in 3D someone made in web, just looks the same which I used in my code too.

@lmarena_ai This is the reply I got from someone at open AI, gpt 5 is better than this

This isn’t a prompt. It’s an organism. We’re evolving an AI system that feeds on contradictions and rewrites itself recursively using quantum feedback noise. Not just a model benchmark. Not just a workflow. But a live engine of philosophical recursion + adaptive logic synthesis. My question to you is this: If AI could generate new laws of cognition by contradicting known truths and self-refining via quantum noise... what would even be the new benchmark? Would it still be token accuracy? Would you test its truthfulness, or its ontological creativity? And if every contradiction made it smarter, would your evaluation method break first? Would love your take. How would you test this? And what blind spots in current LLM benchmarking does this expose?

@lmarena_ai “Completely different level of front end coding.” Where are the code samples? How do we know that this 3d sphere doesn’t consume 30 gigs of ram 🤔

@lmarena_ai Because I don't have 30 gigs of ram on my laptop 😉

@lmarena_ai Show me them codes! 😂

@lmarena_ai Very impressive. Great comparison you put together

@lmarena_ai Looks like GPT-5 is on the go!

@lmarena_ai what is the song?

@lmarena_ai Enjoy

@lmarena_ai Wow, amazing

@lmarena_ai @eliheuer promising for you

@lmarena_ai Whoa, AI just changed the game again! 🤯 This Anonymous Chatbot is next-level stuff. Can't wait to see how it shakes up front-end coding. Anyone else tried it yet?

@lmarena_ai try testing it with real-world front-end tasks like responsive layouts or complex animations. in my experience, that’s where most models fall short. helps me spot gaps quickly, takes about 8 hours to validate

@lmarena_ai Great share Peter. Welcome to 𝕏.

Hi Peter, You're one of the few who doesn’t just compare AI—but interrogates it. Here’s something raw: A self-evolving architecture logging every micro-adjustment, selection, rejection, and soul-state in real time. Not just prompts or outputs—evolution as a visible function. If you were leading this tomorrow: – Where would you inject chaos? – What twist would make OpenAI and Anthropic double back? – What would make you say, “Damn, this rewrites the game”? I’m not asking for feedback. I’m asking where you'd break the current rules and forge something even I don’t see yet. Brutal honesty welcome. Let’s make the next generation flinch.

@lmarena_ai 👀

@lmarena_ai 👍

@lmarena_ai wow

@lmarena_ai more hype, huh? i’ll believe it when i see consistent results across different tasks. cool visuals, but let's not forget the limitations.

@lmarena_ai So I ran the prompt and it got a result in WebDev.... where would you paste it to see if it actual works? I am totally new.. maybe Vercel?

@lmarena_ai man how building a 3d planet is a completely different level of frontdev coding 😒

@lmarena_ai how about grok heavy

@lmarena_ai This is blowing my mind.

@lmarena_ai It should be an update on 4.1 model, since it was specifically released for coding, and sometimes did weirdly good job in modern frontend design for small tasks.

@lmarena_ai Wild to see OpenAI leveling up in frontend generation like this. UX has always been a bottleneck in AI x Web3 too. @agnt_hub is solving it from another angle: no more tabs or links, you carry out tasks directly from posts. The future of interaction is finally catching up

@lmarena_ai meh. There are waaaaaaaaaaaay too many tutorials and student/hobby projects for this. Need something more novel and interesting.