Unverified50% confidenceEventExact time
In February of this year, OpenAI stopped reporting SWE-bench scores for the best-known versions of its flagship models, citing test flaws and training data contamination.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/5/2026
First Seen
Valid until: 1/3/2027
Sources
Related Entities
Related Claims
UnverifiedKapoor and Narayanan criticized OpenAI for disabling known safety guardrails, lacking monitoring in evaluations, and using a testing framework different from the production environment, characterizing it as an organizational failure.67% similarUnverifiedEvaluation benchmarks face data contamination issues, where training data may already include test questions causing inflated scores; in the GPT-4 technical report, OpenAI publicly discussed this issue for the first time.66% similarUnverifiedAnthropic did not report data for many benchmarks in this release, whereas OpenAI provided nearly all metrics.66% similarUnverifiedOpenAI explicitly prohibited using its outputs to train competing models in its early terms of service, and later relaxed this stance somewhat but restrictions remain.63% similarUnverifiedOpenAI曾在2022年内部测试过类似文本水印方案但最终未公开部署，部分原因是对鲁棒性不足的担忧63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/976955API
curl https://kongchang.com/api/v1/knowledge/claims/976955MCP
get_claim(id=976955)