Qwen3.8-27B runs frontier-class coding agents locally

Alibaba's Qwen3.8-27B, a 27.3B-parameter dense model with hybrid attention, ships under Apache 2.0 with a 262,144-token context. It scores 61.7 on SWE-bench Pro and 89.2 on GPQA Diamond, running at ~14 tokens/s on a Mac Studio M3 Ultra via Ollama.
How this story unfolded
4 weeks · 8 reports · 134 community posts · 142 of 150 shown
- Aug 3
- Aug 12
- Aug 13
- Aug 14
- Aug 15
- Aug 16
- Aug 17
- Aug 18
- Aug 19
- Aug 20
- Aug 21
- Aug 22
- Aug 23
- Aug 24
- Aug 25
- Aug 26
- Aug 27
- Aug 28
Qwen by email
Get an email when Qwen has news
No news that day, no email.
More stories today
- MEES, Minimax H3 experiment
- llama.cpp PR list targets faster CPU inference
- Tutorial: Build ensemble weather forecasts with NVIDIA Earth2Studio
- AI band gets YouTube Official Artist Channel status
- Sony Music, Warner sue Anthropic over alleged IP theft