Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I managed to get GPT-5-Codex-Mini to draw me a pelican. It's not a very good one! https://static.simonwillison.net/static/2025/codex-hacking-m...

For comparison, here's GPT-5-Codex (not mini) https://static.simonwillison.net/static/2025/codex-hacking-d... and full GPT-5: https://static.simonwillison.net/static/2025/codex-hacking-g...

I had quite a fun time getting those pelicans though... since GPT-5 Codex Mini isn't officially available via API yet I instead had OpenAI's Codex CLI tool extend itself (in Rust) to add a "codex prompt ..." tool which uses their existing custom auth scheme and backend API, then used that to generate the pelicans. Full details here: https://simonwillison.net/2025/Nov/9/gpt-5-codex-mini/



Can't believe I'm saying this, but GPT-5's pelican is its most impressive improvement over 4o I've seen. I wonder if Codex' MoE does not contain any expert fit for this task due to its fine-tuning on code.


Make sure you check GPT-5.1’s from a couple days ago


wait it's not yet out, isn't it?


It's being tested already on huggingface under another name


Just curious, do you think LLM makers are deliberately adding training data for your pelican test by now?


I'll know if they do, because I'll notice that a new model is suspiciously great at drawing pelicans riding bicycles while still sucking at drawing other things.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: