Deepseek-v3.2

Feature Requests
1

The rational in short:
An up-to-date fast inference non-thinking model that competes with current foundation models.

—-

I’m using the currently different models and I’m not impressed. I will be not able to use them moving forward. We are getting complaints from users about the quality. Groq and Cerebras don’t have the v3.2, so Groq would have an edge by providing it.

Model: https://openrouter.ai/deepseek/deepseek-v3.2

3 Likes
2

I use this model on Deepseek api, and it’s great. I’ve said it before, on creative writing and text analysis Deepseek is the best open source model out there I’ve tested.

1 Like
3

I agree, I’m testing it right now. So far it’s the best open source model I’ve tested. I was using the Qwen3-235B (Instruct) on prod, but I will now exchange it for good and continue with the Deepseek v3.2

1 Like
4

Deepseek-v3.2 is probably the only production-ready open-source model worth hosting on Groq. All the other ones are good for demos or mid user experiences with Kimi K2 at best.