Qwen 3.8-27B Long Thinking Test: Fails to Finish Planning After 50k Tokens
LippyBumblebutt · reddit · 2026-08-17
A user reported performance issues with Qwen 3.8-27B's long thinking mode on a single RX9070 (Q3KS): it failed to finish the planning phase after 50k tokens for a simple CLI task, whereas the previous 3.6 version completed it easily. Adjusting reasoningeffort and KV quantization settings did not resolve the issue.
Related event: Qwen xhigh Mode Consumes Massive Token Count(2 posts)→
More from Models
- GLM 5.5 rumored to drop soon, coding ability expected to match frontier models — bindureddy · 2026-10-16
- Five rounds of adversarial prompting make Opus 5.5 and Argon converge on identical output — LanceCharger · 2026-10-16
- DIY benchmark ranks 18 models on subjective judgments, GLM-5.3 tops consensus match — arkuto · 2026-10-16
- Bindu Reddy: Google's Argon excels at complex doc writing at half the cost of Opus — bindureddy · 2026-10-16
- MAI-2 is 2x bigger than V4-Pro while V4.1F is 3x smaller: Microsoft's scaling ladder questioned — teortaxesTex · 2026-10-16
- Step 5 Preview hits DeepInfra day-zero: 600B MoE, 1M context, $1/M input tokens — StepFun_ai · 2026-10-16