Skip to main content

Update to Qwen3 235B A22B Thinking 2507

Feature RequestsNew Model

https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507

We need more this version that is reasoning, not the Instruct version (without reasoning), as a good replacement to the DeepSeek R1 that was removed from Venice, this new version of Qwen3 235B A22B outperforms in the benchmarks the first version of DeepSeek R1 and even the second one, the 0528.

I think it could even replace DeepSeek R1 in the API.

2 comments

This request was merged into another request

Comments2

  • Black Quasar

    •

    Aug 1, 2025

    This is exactly what I was looking for. The new qwen 235b 2507 is crazy good. I really don’t understand why r1 is also still on this very old version, when we have 0528.

    Is it so difficult to update? I also agree to just let go of llama 3.1 405b to make room for better models. Stuff is quite old here, but especially the qwen 235b should be updated, now that it’s one of the top models

  • Coffee Mug

    •

    Jul 25, 2025

    Llama 3.1 405B Instruct can be removed to give more hardware to this model that outperforms DeepSeek R1 0528.