Unverified50% confidenceBenchmarkExact time
早期Codex在HumanEval基准测试中以37%的准确率独立解决编程题，达到当时最高水平
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
GPT-5.6 Comes to Codex, Voice Models Iterate: A Flurry of AI Product Updates
bilibiliinfinite灵感港7/7/2026
Related Claims
VerifiedCodex achieved a pass@1 accuracy of 28.8% on the HumanEval benchmark81% similarUnverifiedThe tester estimated that Codex can complete approximately 30% of the tasks on the kanban board.73% similarUnverified初版 Codex 采用约120亿参数规模，在 HumanEval 基准测试中达到 28.8% 的 pass@1 通过率72% similarUnverifiedAfter issuing the instruction to Codex — 'No matter how many credits are consumed, after fixing, perform real tests, send real requests to verify whether the API can return results, and keep testing until successful' — Codex ran for 36 minutes and ultimately completed the debugging autonomously.70% similarUnverifiedIn recent hands-on tests, Codex has surpassed Claude Code in multiple evaluation dimensions.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/488600API
curl https://kongchang.com/api/v1/knowledge/claims/488600MCP
get_claim(id=488600)