Qwen 3.8 default temp 1.0 causes garbage output, 0.1 fixes it
KeepyUpper · reddit · 2026-08-19
A user reported that Qwen 3.8 produces garbage output—hallucinations, cut-off sentences, or character spam—after 10-20k tokens when using the recommended high temperature (1.0).
Issue:
- Occurs on both vLLM and llama.cpp.
- The thinking process breaks down and devolves into nonsense.
Fix:
Lowering the temperature to 0.1 immediately resolves the issue and also improves generation speed. The user suspects Qwen 3.8 might be particularly sensitive to KV cache quantization or requires specific configuration changes, as Qwen 3.6 did not exhibit this behavior.
Related event: Qwen 3.8 Default Temperature Causes Gibberish, Lower It(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