Unverified50% confidenceEventExact time
A Reddit user claimed that a locally run Qwen3.8-27B model (Q4 quantized, run via llama.cpp on an RTX 4090) could handle complex tasks and bug fixes early after release, but recently it frequently errors even on basic tasks.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
11/5/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 11/19/2026(expired)
Sources
Has Qwen3.8-27B Become "Dumber"? Users Suspect the Model Was Secretly Weakened
redditr/LocalLLaMA11/5/2026
Related Entities
Related Claims
UnverifiedRunning `heretic Qwen/Qwen3.5-4B` completes a basic uncensoring operation.72% similarUnverifiedA developer ran the Qwen3 27B model locally on a single RTX 5090 graphics card and shared the test results.72% similarUnverifiedA project called Qwenputer embeds the behavior of a 32-bit RISC-V CPU into the weights of the Qwen3.5-4B model, enabling the model to execute RISC-V instructions rather than calling an external emulator.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1138696API
curl https://kongchang.com/api/v1/knowledge/claims/1138696MCP
get_claim(id=1138696)