I'd also argue this is the case for any company releasing open weights. They're not righteous, they're marketing. That's not necessarily a bad thing! They're releasing some great stuff for free and we benefit from that. Every company doing this has a motivation to not release these for free.
Alibaba, Google, Moonshot, Thinking Machines, etc are not releasing their models for free because they love to. They want to grab market share. I'll take it.
I still will not use a hosted Meta product, but damn this model looks solid.
{"deleted":true,"id":49243567,"parent":49243445,"time":1786369429,"type":"comment"}
This model doesn’t look solid at all. It comes months after the Qwen model, and in almost half the benchmarks, it performs worse than that. Plus, the next Qwen 3.8 is going to be announced this week. So, this model is DOA.
I don't really care that much about benchmarks, but having tested it on one of my puzzle prompts I can tell you that it solves it well, writes clearly, isn't noticeably slower than Qwen 3.6 27B and is much more terse in its reasoning (which will help with preserve-reasoning).
It also has a knowledge cutoff inside this year.
The main limitation is the smaller maximum recommended context.