Hacker News new | past | comments | ask | show | jobs | submit | elvennn's comments login

Indeed it does. But also the price for output tokens of the OCR is cheaper. So in total it's still much cheaper with gpt-4o-mini.


Regarding Mistral recent announcements, I'm not so sure anymore.


The large models can't be run by consumers, and why would a for profit company release a large model that can then be picked up by other startups?


They could re-use the Llama license but with a lower threshold (say 50000 instead of 500M) and call it a day, but they decided not to.


Where can I download the 70b mistral medium then? Other than the miqu leak.


21k LOC ...


yea 21k lines of code is like a tinymicro project :|


One of Shiro Games founder (Nicolas Cannasse) is the creator of the Haxe language and Heaps engine !


I think the game is Dofus, a French MMO ;)


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

Search: