From their post:
Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window.
Highlights:
- Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation.
- Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables.
- Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery.
More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
Same here. The small MoE ones are the only reason local is useful for me at all. I run qwen3:30b-a3b through Ollama on a MacBook and get roughly 70–95 tok/s, which is fast enough that I hand it all the boring jobs: summarising logs, pulling fields out of emails, first-pass triage. Anything that needs judgement still goes to a bigger model, because the a3b will happily do a tidy job of the wrong thing if the instruction is even slightly ambiguous. So I spot-check everything it does.
The other half of my setup is an old i7 home server with no usable GPU, and the same model is close to useless there. So what I’m really waiting for from Qwen4 is a small model that’s smarter per active parameter, not a bigger one. If they ship a ~30B-A3B-class Qwen4 that follows instructions better, that’s an instant swap for me. At least until I can spring for the Strix Halo!
I switched from Gemma4:31b (
UD-Q4_K_XL) to Qwen3.8:27b (same quant) about 2 weeks ago, and I’ve been very impressed with the reasoning abilities, and generally non-sycophantic responses I’m getting from Qwen3, as compared to Gemma. Looking forward to see what improvements v4 brings.I’ve been doing Qwen 3.8 Flash through Strata and it is another step up from 27b. I don’t quite have as much context headroom, but in my experience it uses that context much more efficiently.
Oh yeah, 27B was a massive leap forward for local AI.
deleted by creator
Both Gemma4 31b and Qwen3.x 27b are dense models. I think the difference is mostly about the model architecture and training data.
Gemma4 26b a4b and Qwen 35b a3b are the MoE models.
Btw, I personally like to use the Gemma4 models to help with writing emails etc, whereas I use Qwen3.x models to help me program. I think they are good at those respective roles.
It’s very annoying to receive LLM emails. Please consider not doing that.
Most of my prompts are “tell him to fuck off, but keep it professional”
lol, should try this sometime, just to get a good laugh
I agree. I personally never generate entire emails. However, I sometimes use LLMs to give me alternative ways of saying things or to catch spelling/grammatical errors.
just got confused b/w 26b and 31b
Handles real tasks, including…manufacturing…healthcare…
Can’t wait to see how it does this 😅
It gains control of the missile launch control systems and threatens people to do its bidding unless they want to go bye-bye.
I really need a successor to 35b-a3b!
Qwhen?
Next Qweek maybe?
would be cool to get an upgrade to qwen3.5-9b
2.4 trillion wow
Did they ever say anything about removing the cyber security and reverse engineering capabilities on their smaller models (27b and Flash)?
That could end up making 3.8 more useful than 4 in those areas.
I didn’t see any news about it, but even if they add guards for it - we can remove any guards because these models are open weight.







