Unverified50% confidenceFactExact time
After training for most of a day using only about 1,000 edge tasks that are 'sometimes done right and sometimes done wrong,' the 4B model recovered more than half the gap to the 27B model.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the experiments, the 4B model's accuracy before training was about fifty percent, while the 27B model reached over seventy-five percent.76% similarUnverifiedThe same model under different Harnesses can differ in task completion rate by several times.69% similarUnverifiedA 35B model does not necessarily run faster than a 27B model, and if swap memory is used frequently, the experience may be worse than a smaller model.69% similarUnverifiedFor the same model on the same benchmark, results from different teams can differ by 28 percentage points, and this gap is mainly caused by differences in test environment, prompt design, tool integration approach, and retry strategy, rather than fluctuations in model capability.68% similarVerifiedFor models with more than 30B parameters, the performance loss caused by 4-bit quantization typically results in a 1–3% drop on benchmarks, but the impact is more significant for smaller models.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1056466API
curl https://kongchang.com/api/v1/knowledge/claims/1056466MCP
get_claim(id=1056466)