Unverified50% confidenceFactExact time
In 2023, multiple studies revealed a significant gap between some models' true capabilities and their reported scores on the GSM8K mathematics benchmark.
1
Sources
50%
Confidence
Long-term
Relevance
9/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSelf-Consistency can bring improvements of 5–20 percentage points on mathematical reasoning benchmarks such as GSM8K.77% similarUnverified研究表明，即使在 GSM8K 或 MATH 数据集上接近满分的模型，面对稍作变形或需要跨领域组合的问题时性能下降可达30-50个百分点76% similarUnverifiedThe optimal variant showed only a 1.83% decline on the GSM8K mathematical reasoning benchmark, while the MMLU general knowledge benchmark remained unchanged.73% similarUnverifiedThe gap between all uncensored model variants and the base model on the GSM8K mathematical reasoning test is within 1.2 percentage points.73% similarUnverified在GSM8K小学数学应用题基准上，前沿模型的准确率已超过90%67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/882905API
curl https://kongchang.com/api/v1/knowledge/claims/882905MCP
get_claim(id=882905)