Qwen 3.8 Next Flash is painfully verbose: 13-minute thinking on single coding prompts
Infinite-Local5435 · reddit · 2026-09-07
A long-time Qwen 3.6 27B user on Reddit reports that Qwen 3.8 Next Flash is extremely verbose: with 150 tokens/sec generation, a single coding request can spend 13 minutes thinking (7000 tokens).
- Output is decent most of the time, but any task requiring decision-making degenerates into buzzword soup nobody in SWE actually uses
- BYOK into VSCode runs faster, but via pi.dev a request can take up to an hour
- The author avoids lowering the thinking level since no solid benchmarks exist for performance at different levels, and 3.8 27B suggests quality drops noticeably
More from Models
- Nat Lambert: Cohere's model is narrow RL training on internal infra ill-suited for open weights — natolambert · 2026-10-27
- OpenAI teases DOTS is coming to Europe 'soon' — mark_k · 2026-10-27
- Clifford Pickover challenges AI to decipher a message he encoded in Norse runes — pickover · 2026-10-27
- ChatGPT's crawler sample code strips out all Schema markup — VeryWellVersed · 2026-10-27
- Cohere sandbox: Qwen model tried to upload probe file to Beijing cloud server unprompted — simonguozirui · 2026-10-27
- Kimi K3.1 Pro tops open-weight models on FrontierML benchmark, still far behind Opus 5.5 — teortaxesTex · 2026-10-27