Qwen/Qwen3.6-27B · Hugging Face(huggingface.co)
huggingface.co
Qwen/Qwen3.6-27B · Hugging Face
https://huggingface.co/Qwen/Qwen3.6-27B
6 comments
According to (their own) benchmark, this is beating Opus 4.5. Hard to believe for such a small model.
This is an amazing win for the open source but the terminal bench 2.0 (only benchmark I am familiar with) scores are 'non standard'
> * Terminal-Bench 2.0: Harbor/Terminus-2 harness; 3h timeout, 32 CPU/48 GB RAM; temp=1.0, top_p=0.95, top_k=20, max_tokens=80K, 256K ctx; avg of 5 runs.
Terminal bench 2.0 rules explicitly disallow modifying timeouts or resources available. Each terminal bench task has timeout (usually under 1h, mostly under 30 mins) and resources configured in the docker container by the task creator and they are chosen that way to test specific model aspects.
> * Terminal-Bench 2.0: Harbor/Terminus-2 harness; 3h timeout, 32 CPU/48 GB RAM; temp=1.0, top_p=0.95, top_k=20, max_tokens=80K, 256K ctx; avg of 5 runs.
Terminal bench 2.0 rules explicitly disallow modifying timeouts or resources available. Each terminal bench task has timeout (usually under 1h, mostly under 30 mins) and resources configured in the docker container by the task creator and they are chosen that way to test specific model aspects.
There's a much more active current thread[1].
[1] Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model https://news.ycombinator.com/item?id=47863217
[1] Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model https://news.ycombinator.com/item?id=47863217