Qwen 3.8 27B

	Qwen 3.8 27B(huggingface.co)
	1423 points by erdaltoprak 7 days ago \| 790 comments
	tl;dr: Qwen has released Qwen3.8-27B, a 27B-parameter dense vision-language model with a hybrid Gated DeltaNet/Gated Attention architecture, native 262K context (extensible to 1M via YaRN), and FP8 quantization. It features toggleable thinking mode with tunable reasoning_effort, preserved thinking across turns, and native image/video understanding. Benchmarks claim it outperforms prior Qwen models and competes with Opus 4.6 Max on coding (SWE-bench Pro, Terminal Bench) and agentic tasks, while trailing on some reasoning benchmarks like HLE.
	HN Discussion: ~Model successfully passes personal benchmarks but has efficiency tradeoffs like high VRAM and token usage ↑Impressed with quality of output for a locally-runnable model, praising specific creative generation results ↓Model overthinks and second-guesses, making it less practical than competing efficient models like Gemma ↑Benchmarks approaching frontier closed models suggests local models will soon match top-tier capability ↑Small dense open-weight models like this benefit the public most by being widely accessible