Same prompt. very different execution.
Claude Opus 5 spent almost 10x more and autonomously kept expanding the project beyond the original ask. if models start doing more work without needing more instructions, the gap between “AI tool” and “AI agent” gets very real.
Benchmaxing is real but the directional signal here matters.
Fable 5.1 matching Opus 5 on accuracy with lower latency is a genuine shift in the Pareto frontier.
You have reached the end of the archive
All of Claude Opus 5