REDDIT · REDDIT// 3h agoINFRASTRUCTURE

Qwen3.6-Plus reasoning tokens break OpenWebUI output

Users are reporting parsing issues with Qwen 3.6's reasoning tokens when hosted on LM Studio and used via OpenWebUI. Quotes inside thinking blocks cause the system to prematurely treat reasoning as regular output, breaking tool calls and truncating responses.

// ANALYSIS

This highlights the fragility of local model orchestration stacks as reasoning models become more complex.

–The bug appears tied to how OpenWebUI parses quotes within the thinking tags emitted by Qwen 3.6.
–This issue reportedly did not exist in Qwen 3.5, suggesting a change in the new model's output formatting.
–Broken tool calls point to downstream pipeline failures when reasoning traces bleed into the final response payload.

// TAGS

qwen3-6-plusopenwebuilm-studioreasoninginferencellmopen-weights

DISCOVERED

3h ago

2026-04-18

PUBLISHED

3h ago

2026-04-18

RELEVANCE

6/ 10

AUTHOR

luckyj