I wrote this on
MiniMax M3
The new king of open weights models?
At the beginning of this month, MiniMax released M3. Then, a few days ago, they cut its price by half (perhaps because of Kimi K2.7-Code, or GLM 5.2?). The model is about as smart as the new Gemini 3.5 Flash, but is five times cheaper. You can download its weights via OpenHands, and even run a quantized version of it if you happen to have many hundreds of GBs of RAM. 💸💸💸💸
Unlike Deepseek V4, it is natively multimodal. Many providers already host it. It’s easy to understand why it recently surged to the 2nd most used model on OpenRouter.
In any case, having such a smart model available to download for free is wild. Thank you, MiniMax!
Rate this post 👇 Privacy-first reactions via Reaction Count