DeepSeek-V3-0324 Crushes GPT-4.5 in Math and Code Benchmarks at 1/277 the Cost(api-docs.deepseek.com)

10 pointsby kong75l’année dernière2 comments

https://api-docs.deepseek.com/news/news250325

2 comments

kridsdale1l’année dernière

Well yeah. 4.5 is not a reasoning model. It’s a one shot “creative” model. Nobody expects coding or math from these anymore.

syntaxingl’année dernière

For what it’s worth, V3 is not a reasoning model either. Chances are, this V3 checkpoint is probably a distill of the not yet released R2.