qwen3-6-35b

By pairing Qwen3.6-35B with the little-coder agent scaffold, the model hits 78.7% on the Polyglot coding benchmark, placing in the public top 10 and rivaling cloud models.