The team say it beats 3.5 27B in numerous benchmarks. But it's still a MoE with 3B active per token, so we'll see!
I hope there's a 3.6 27B coming in short order, or even a 31B.
Gemma 4 looked great, but I've not got it working well with Claude code (Qwen just works) - Gemma 4 _does_ work with opencode though.
alexellisuk•1h ago
The team say it beats 3.5 27B in numerous benchmarks. But it's still a MoE with 3B active per token, so we'll see!
I hope there's a 3.6 27B coming in short order, or even a 31B.
Gemma 4 looked great, but I've not got it working well with Claude code (Qwen just works) - Gemma 4 _does_ work with opencode though.