FULL STORY

Qwen 3.8 27B: From Release to Community Benchmarks

Days after Qwen 3.8 27B shipped, community testers found it beating the prior 3.6 in bf16 web-design tasks and out-coding 30B-class rivals like Gemma 4 on local GPUs.

2026-08-14 ~ 2026-08-16 · 2 episodes · 6 posts

Episode 1 · Qwen 3.8 27B Released as Community Tests Performance and Speed (2026-08-14, 4 posts)

Qwen 3.8 27B has drawn strong community feedback, with users reporting better web design output than 3.6 at bf16 and about 22 tokens/s using q4 quantization with MTP, while debating whether the 27B model can beat the 122B Qwen 3.5 and how close it is to frontier models.

Episode 2 · Qwen2.5-32B Beats Qwen2 and Gemma 4 in Coding, Agentic Benchmarks (2026-08-14, 2 posts)

Users benchmarked Qwen2.5-32B (dubbed 'Qwen3.8' in posts) against Qwen2-32B and Gemma 4 locally on a 24GB GPU, finding Qwen2.5 leading on coding and agentic benchmarks among the similarly sized open models.