Post

New super-fast model from OpenAI today powered by their new Cerebras partnership - GPT-5.3-Codex-Spark It's 4-5x faster than GPT-5.3-Codex but the pelican isn't as good! Here's its pelican compared to full GPT-5.3 Codex, both on "medium" simonwillison.net/2026/Feb/12/...

12/2/26, 9:24 pm6 reposts74 likes

eastmad

@eastmad.bsky.social

12 Feb

I'm sorry, I can't unsee the One True Pelican of Gemini Deep Think. These are unworthy

Joseph Humfrey

@joe.inkle.co

13 Feb

Have you got a library of old pelicans you can scan through to check what it’s closest to? 4o, 4.1 mini, etc?

Simon Willison

@simonwillison.net

13 Feb

Mainly on my tag page simonwillison.net/tags/pelican...

simonwillison.netSimon Willison on pelican-riding-a-bicycle94 posts tagged ‘pelican-riding-a-bicycle’. My benchmark for LLMs: "Generate an SVG of a pelican riding a bicycle". "User might be a kid playing with words" according to Qwen3-4B-Thinking.

Joseph Humfrey

@joe.inkle.co

13 Feb

Haha amazing tag, love it, and it’s fascinating. Thanks!

Jared Nicholson

@jarednicholson.bsky.social

12 Feb

Interesting, I did think the mistral le chat flash answers were worse the slow ones.

Mike

@mikeschroeder.bsky.social

13 Feb

Not as good? That pelican is abysmal!

Chris Ozols

@twolongos.com

12 Feb

I don't understand the appeal of "fast" coding models. I don't care when its done I just want it done right. If the fast ones could do it both right and fast, it would just be the base model. What's a use case for the Max Power coding agent? "You'll get to code review faster"?

Simon Willison

@simonwillison.net

12 Feb

It means you can iterate with them at real-time speeds which is pretty fun simonwillison.net/2024/Oct/31/...

jack

@j4ck.xyz

12 Feb

I’m pretty excited to have the choice for quick models tho, it feels soooooo slow to me normally (even in-app GPT models)