Expired50% confidenceBenchmarkExact time
The earliest models scored less than 20% on the GDPVal benchmark, far below human performance.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/9/2026
First Seen
Valid until: 10/7/2026(expired)
Sources
OpenAI's Head of Evals: We've Been Consistently Underestimating What AI Models Can Really Do
bilibiliGelai_AI7/8/2026
Related Claims
Unverified在SWE-bench Full版本上，多数模型的通过率仍低于20%72% similarUnverifiedTraditional inference solutions often achieve GPU memory utilization of only 20%–40%.67% similarUnverifiedSome research estimates that fewer than 20% of strategies selected through backtesting within quantitative funds remain consistently effective in live trading.66% similarUnverified2024年多项研究发现，部分模型在去污染版本的评测集上得分下降幅度可达10-20个百分点65% similarVerifiedIn early 2024, the strongest AI models scored only in the 10%–20% range on SWE-Bench.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/362880API
curl https://kongchang.com/api/v1/knowledge/claims/362880MCP
get_claim(id=362880)