Repository Issue Activity (beta)

#212 [Bug] `Ask this replies that` issue persisted

#214 Qwen3.8-27B: premature EOS right after a backtick in long-context thinking mode (SGLang with EAGLE spec decode)

#215 Qwen3.8-27B consistently outputs left_click action despite SYSTEM_PROMPT restriction in mobile GUI environment.

#239 A little ai hallucination at qwen 3.8 max

#179 Reproducing Qwen3.6-27B SWE-bench Pro (53.5): a `str_replace` file-edit tool + reliable-test conditioning closes the gap (bash-only ~28% → ~51% pass@1 / ~71% pass@8)

#208 Content filter blocking legitimate FNF character discussions

#220 [Badcase] Qwen3.6-27B BF16 drops 虑 from 深思熟虑 during sampled translation with vLLM

#195 web UI/UX becomes worse, after last update

#192 about qwen3.7-Plus，Why is the error rate for such a simple question so high? (To the Qianwen R&D team)

#204 Qwen-3.8-Max: Updated files with the same name are ignored due to context/session caching

#203 Profile-based MoE routing: making Qwen's experts swappable post-training?

#238 Tamil (agglutinative-language) quality regression vs Qwen3.6-27B, not fixed by better DeltaNet quantization protection

#237 Tamil (agglutinative-language) quality regression vs Qwen3.6-27B, not fixed by better DeltaNet quantization protection

#236 [Bug] Tool calling becomes unreliable with long prompts when enable_thinking=false (Qwen3.8-27B/Qwen3.8-27B-FP8)

#235 Ask with chinese reply with japanese

#145 Qwen3.5 series models easily fall into infinite loops during reasoning with the provided sampling parameters

#216 Qwen3.8-27B: reasoning_effort="xhigh" (the default) returns an empty answer with finish_reason="stop" on ~17% of calls

#231 Qwen3.8-27B unexpectedly truncates output after emitting “畅皓”是我武生物的黄

#230 Frequent interruptions when using Claude Code with Qwen3.8 Flash

#229 GUI Tasks System Prompt or Cooking Script

#187 Qwen3 XML tool-call format is not fully type-safe under JSON inputs

#186 Incompatibility with Qwen3.5/3.6 models caused by mid-conversation system messages adaptation(After claudecode v154)

#226 [Bug] Qwen Desktop (Windows): next-turn context drops previous turn's tool_calls/tool_results → model denies its own tool executions

#217 Qwen3.8-27B chat template rejects Claude Code's default reasoning_effort="high" via vLLM → HTTP 500

#223 Which framework/harness is used for vision benchmarks reported "With CI" (e.g., BabyVision)?