Unverified50% confidenceBenchmarkExact time
审查结果显示 Opus 5 频繁杀错进程（常误杀自己所在的 T3 Code 实例），Sol（GPT）有约40%概率提交 draft PR
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
8/22/2026
First Seen
Valid until: 9/5/2026
Sources
Related Entities
Related Claims
UnverifiedOpus 4.7每次解题的中位输出token数为60,000，而GPT 5.5仅需16,00072% similarUnverified在Agents' Last Exam评测中，GPT-5.6 Sol取得53.6分，比采用自适应推理的Claude Fable 5高出13.1分72% similarUnverifiedClaude Opus 4.8在AI Coding基准测试中表现超越GPT-5.570% similarUnverifiedOpus 4.7在长上下文（50万-100万token）检索方面表现明显退步，远不如DeepSeek和GPT 5.567% similarUnverified据非官方爆料，Terminal Bench 3.0基准测试中Claude Opus 5 Max以43.5%的得分超越GPT-5.6-Sol的34.6%重登榜首67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/788387API
curl https://kongchang.com/api/v1/knowledge/claims/788387MCP
get_claim(id=788387)