Unverified50% confidenceFactExact time
When mainstream models' accuracy on GSM8K approaches 95% or above, the benchmark loses its discriminative power.
1
Sources
50%
Confidence
Long-term
Relevance
10/29/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn 2023, multiple studies revealed a significant gap between some models' true capabilities and their reported scores on the GSM8K mathematics benchmark.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75% similarUnverifiedSelf-Consistency can bring improvements of 5–20 percentage points on mathematical reasoning benchmarks such as GSM8K.74% similarUnverifiedOn a real LLM (DeepSeek), the researchers evaluated 50 samples each on the GSM8K, MATH, and BBH benchmarks, and GraphEvoResearcher achieved accuracies of 98%, 92%, and 82% respectively, the highest among all baselines.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%73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1110820API
curl https://kongchang.com/api/v1/knowledge/claims/1110820MCP
get_claim(id=1110820)