Qwen 3.7 Plus has been released

Posted by WhyLifeIs4@reddit | LocalLLaMA | View on Reddit | 19 comments

https://qwen.ai/blog?id=qwen3.7-plus

[-]

thecratedigger_25@reddit

I have some Qwen models on my pc that run 7b parameters locally and it works well. Can't wait to see what the new Qwen will do.

[-]

SDGANON@reddit

I'm running Qwen3.6-35B moe and getting \~50t/s on my 5080 along with Open WebUI.

So far I've been pretty happy with it. I tried 2.5 at a slightly smaller size but 3.6 is clearly better.
Really hoping we get a 3.7 weight release soon.

[-]

kevinlch@reddit

They cherry-picked weaker model for the benchmark result. Do you guys still use GPT-5.4? Not cool.

[-]

effortless-switch@reddit

Sure but 5.4 was like SOTA according to many just a few weeks ago. I'll take it.

[-]

ttkciar@reddit

Until such time that weights are published, this is off-topic for LocalLLaMA.

[-]

FullstackSensei@reddit

Call me when the weights are released.

Otherwise, API or whatever isn't interesting

[-]

GreenGreasyGreasels@reddit

It will probably not be open sourced. I think the Qwen pattern is release a instruct model checkpoint with a little fine tune on top. This is the open source model.

Then it iterate on the same model with post training and enhanced harness. Sell this as the plus module. I guess that way they think they contribute to open source while still retaining a s edge for commercial activities.

I wish it wasn't this way, but I think I understand why they do it.

[-]

Finanzamt_Endgegner@reddit

Was available before in qwen chat no?

[-]

NickCanCode@reddit

That was the MAX version I believe.

[-]

Finanzamt_Endgegner@reddit

no i thin the plus was there but only max was officially released, meaning there werent any benchmarks or announcements, was already thinking it was weird that they released 2 models but only announed 2

[-]