MiMo v2.5 is underrated. Feels like the tokens are pouring out of the screen in OpenCode.

When I recently built my inference server, I expected to deploy DeepSeek v4 flash, but that doesn't look like it's going to be fast for a long time, if ever. There is a massive gap, as we all know, in competent models between 30b and 400b. I was very surprised to find that this is the best mod

中文全文翻译正在补齐，可先查看下方英文原文。

When I recently built my inference server, I expected to deploy DeepSeek v4 flash, but that doesn't look like it's going to be fast for a long time, if ever. There is a massive gap, as we all know, in competent models between 30b and 400b. I was very surprised to find that this is the best mod