8.1
Qwen 3.8 27B Review — A 27B Dense Model That Beats Opus 4.6 Max on Agentic Coding Benchmarks
Qwen3.8-27B is the compact flagship of Qwen's open-model family: a 27B dense vision-language model with 262K native context (extensible to 1M), thinking control via reasoning_effort, and benchmark scores that punch far above its weight — SWE-bench Pro 61.7, Terminal Bench 2.1 (Terminus) 73.0, DeepSWE 1.1 42.2 (beating Opus 4.7 Max's 40), OSWorld-Verified 84.3, WebArena-Verified 64.8. FP8 weights ship on Hugging Face (Apache-2.0) with Unsloth GGUF/NVFP4 quants. Review covers benchmarks, the hybrid DeltaNet architecture, deployment reality on consumer hardware, and the HN debate on whether small models really 'beat Opus.'