MiniMax M3 is surprisingly powerful, and open weight (or is about to be). There's others in this space too: MiMo v2.5, GLM 5.1. There's quite a few to pick from if you want strong models running on "your" hardware.
Fun fact, I was trying this afternoon Deepseek vs Opus 4.8 high, and I was surprised at how good Deepseek was. It outperformed Opus 4.8 on multiple occasions.
Found just later I was using v4 flash and not pro (for mistakenly setting the model to deepseek-chat and not v4-pro).
There are aspects about Deepseek I don't like though, when pushed against it will eagerly bend instead of reasoning and advocating for his points, something Opus 4.7 and later models started doing a lot (even when wrong).
MiniMax M3 is surprisingly powerful, and open weight (or is about to be). There's others in this space too: MiMo v2.5, GLM 5.1. There's quite a few to pick from if you want strong models running on "your" hardware.
MiniMax M3 weights have already been released: https://huggingface.co/MiniMaxAI/MiniMax-M3
Open weights like this make me wish I had a bunch more DGX Sparks to cluster so I could fit it!
US is well known to impose world wide embargoes on technologies and resources that they pretend to apply to companies beyond its borders.
And it works, because the American market is generally more important than the markets of these countries 9 times out of 10.
deepseek v4 pro is great and open weight.
It is, and I love it, but it isn't capable of performing the tasks I've been giving to Opus, let alone Fable.
Don't get me wrong, I use it, it's fast-smart-and affordable. But not suitable for all tasks.
What kinds of tasks are you finding deepseek v4 incapable of?
2 replies →
Fun fact, I was trying this afternoon Deepseek vs Opus 4.8 high, and I was surprised at how good Deepseek was. It outperformed Opus 4.8 on multiple occasions.
Found just later I was using v4 flash and not pro (for mistakenly setting the model to deepseek-chat and not v4-pro).
There are aspects about Deepseek I don't like though, when pushed against it will eagerly bend instead of reasoning and advocating for his points, something Opus 4.7 and later models started doing a lot (even when wrong).
All current Qwen 3.7 models are closed though they have said more releases are coming