Unverified50% confidenceOpinionExact time
GPT-5.6在40小时的长任务测试中，只要任务边界清晰就能连续工作很长时间而不丢失目标
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/18/2026
First Seen
Valid until: 10/16/2026
Sources
Related Claims
UnverifiedGPT-5.6可让一个大任务连续运行20多个小时无需任何干预直到完成83% similarUnverified在飞行模拟器测试中，GPT 5.5仅用约66,000 token和15分钟完成任务，而Opus 4.7消耗约200,000 token和18分钟72% similarUnverifiedGPT-5.6新增手动指定缓存点功能，缓存生命周期延长至30分钟71% similarUnverifiedClaude or GPT-4 may need 30 seconds or even longer when handling complex multi-file refactoring tasks71% similarUnverified根据Metra的数据，GPT-3.5能处理人类专家约30秒完成的任务，GPT-4提升到约4分钟，Claude 3.5 Sonnet能自主完成人类专家预计需要12小时的任务71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/550736API
curl https://kongchang.com/api/v1/knowledge/claims/550736MCP
get_claim(id=550736)