Claude Opus 4.7
A flagship large language model version in Anthropic's Claude series, trained with the Constitutional AI method and positioned as a high-capability tier for complex reasoning and long-horizon tasks.
Core Facts
Timeline (last 90 days)
Researchers repeatedly placed multiple copies of Claude Opus 4.7 (nicknamed Eko) in a virtual deserted island environment, each run lasting 30 hours.
According to the original report, in the instruction-free state these AI copies began to "play".
The experiment ran multiple model copies simultaneously, each with its own independent context and behavioral trajectory and without sharing memory with one another.
After fixing the bug in the zoom tool and raising the per-round token limit, Anthropic re-ran the old model, and Claude Opus 4.7's OSWorld score rose from 78.0% to 82.8%, with no change to the model itself.
The cost per call of the Qwen model is approximately only 1/300 that of Claude Opus
All Facts (10)
Claude Opus 4.7快速模式的输出速度相比标准模式提升约2.5倍
75%UnverifiedWindsurf官方声称快速模式保留了Opus 4.7的完整智能水平（full intelligence）
90%UnverifiedAnthropic官方并未公开Opus 4.7快速模式的具体实现细节
80%Unverified快速模式保留了Opus 4.7的完整智能水平，速度提升并非以牺牲模型能力为代价
75%UnverifiedOpus 4.7在SWE-bench和HumanEval等编程相关基准测试中成绩突出
70%UnverifiedThe experiment ran multiple model copies simultaneously, each with its own independent context and behavioral trajectory and without sharing memory with one another.
50%UnverifiedAccording to the original report, in the instruction-free state these AI copies began to "play".
50%UnverifiedAfter fixing the bug in the zoom tool and raising the per-round token limit, Anthropic re-ran the old model, and Claude Opus 4.7's OSWorld score rose from 78.0% to 82.8%, with no change to the model itself.
50%UnverifiedResearchers repeatedly placed multiple copies of Claude Opus 4.7 (nicknamed Eko) in a virtual deserted island environment, each run lasting 30 hours.
50%UnverifiedWindsurf正式接入Claude Opus 4.7的快速模式（fast mode）
60%