Radar Ultra

Model

X @arena

Arena.ai says Claude Fable 5.2 has a higher no-correction response rate than Claude Opus 5.5 on longer tasks

Arena.ai's multi-turn session data indicate that about 85% of responses from both Claude Fable 5.2 and Claude Opus 5.5 needed no correction in tasks with up to 3 replies. For tasks with 8–15 replies, the rates were 89% and 83%, respectively; for tasks with 16 or more replies, they were 93% and 85%.

Arena.ai also said users built directly on Claude Fable 5.2's first response 82% of the time, compared with 67% for Claude Opus 5.5; users pushed back on the models' first responses 13% and 27% of the time, respectively.

Written by AI from the reports · updated 5 days ago

Sources

Radar Ultra
An unofficial tool, not affiliated with OpenAI. Compiled from public posts; the original posts are authoritative.