Browser UseFrom the builder
Claude Fable Sets High Score on BU Bench
Browser Use tested Anthropic's new Claude Fable 5 with its open-source library, and the model scored 80.0% on BU Bench V1, 12 points ahead of second-place GPT 5.5. Fable passed 80 of 100 tasks at an average of 6 minutes 53 seconds each, but the run cost $580.87 in API fees. Of the 32 tasks GPT 5.5 failed, Fable passed 23, and its own failures mostly came from wrong facts or entities, poor source access, or answers the trace did not fully support.
The result came at an exorbitant price: $580.87 in API cost for 100 tasks.
Saved . A shortened excerpt from the official blog; the original publisher retains authorship.