Unverified50% confidenceBenchmarkExact time
In the experiment, Level 3 (full checklist prompt) scored an average of 30.3/40 (76%), higher than Level 1's 24.7/40 (62%) and Level 2's 22.2/40 (55%).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedLe Chonk scored a total of 42/80, or 52.5%, equivalent to 5.25/10, across the eight questions in KingBench3.66% similarUnverifiedThe current top model's comprehensive test score is 58, 24 points above Qianwen 3-27B's 34.64% similarUnverifiedThe author uses a weighted scoring scheme: in the outcome dimension, Pass scores 1.0, Partial scores 0.5, and Fail scores 0; severity weights are 4 for Critical, 3 for High, 2 for Medium, and 1 for Low.61% similarUnverifiedActual testing of two quantization models: IQ2_XXS achieves 48 tokens/s for short Q&A and 63 tokens/s for long text; IQ3_XXS achieves 30 tokens/s for short Q&A and 39 tokens/s for long text, with VRAM usage of approximately 16.44 GB, activating 10,138 experts—IQ2 is about 60% faster than IQ3.61% similarUnverifiedThe relevance and coverage of scoring rubrics generated by CriticGen improved from 3.33/4.03 to 3.97/4.24, respectively.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1043899API
curl https://kongchang.com/api/v1/knowledge/claims/1043899MCP
get_claim(id=1043899)