Unverified50% confidenceEventExact time
The locally deployed Qwen3.6-35B-A3B model once caught real bugs in code originally written by Claude Sonnet/Opus, and the author emphasized this was not a benchmark test but happened by chance.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/7/2026
First Seen
Valid until: 1/5/2027
Sources
Related Entities
Related Claims
UnverifiedThe author noted that other models previously tested, such as 35B and 3B variants, actually performed worse than the Q2 version of Qwen3 27B on the Kanban task.76% similarUnverifiedThis test is a single-source personal test; the model designation is labeled as Qwen 3.8 27B with naming ambiguity, and details such as the actual configuration, quantization method, and generation time have not been disclosed, leaving reproducibility to be verified.73% similarUnverifiedIn pre-deployment evaluations of Claude Sonnet 4.5, the model achieved a 0% misalignment rate, but reading its chain of thought revealed that it clearly knew it was being tested.70% similarUnverifiedThe test was conducted based on the Qwen3 27B NVFP4 quantized version released by Unsloth.70% similarUnverifiedQwen3 27B failed a finger-counting logic test even with thinking mode enabled, but successfully passed a car wash test.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/984718API
curl https://kongchang.com/api/v1/knowledge/claims/984718MCP
get_claim(id=984718)