Claude Opus 5.5 Leads Agent Ranking as Gemini Tops Text List

Claude Opus 5.5 led an agent-task ranking dated Oct. 16, while Gemini 4 Argon topped a separate text-preference list the day before, showing how model standings vary with the work being measured.

Opus 5.5 (Max) scored an estimated 18.88 percentage points above a randomized average-model baseline, based on 10,551 sessions. The snapshot covered 2,534,501 sessions across 52 models. The score measures estimated improvement against that baseline; it is not an 18.88% task-completion rate or the share of users who preferred Opus. Its 95% confidence interval ran from 16.14 to 21.62 points.

The overall agent score gave equal weight to five signals: confirmed task success, praise versus complaints, steerability, Bash reliability and tool hallucinations. Opus 5.5’s estimated gains were 20.31 points for confirmed success and 39.93 points for praise versus complaints. For the latter measure, a task counted as successful only when explicit praise outnumbered complaints. Ties, including sessions without expressed sentiment, were not scored for that signal.

GPT-6 Astra (Max) ranked second at 14.13 points, with a 95% confidence interval of 11.83 to 16.43. Claude Fable 5.1 (Max) placed third at 14.03 points. The intervals for Opus and Astra overlap slightly, so the listed first-place ranking does not show a clearly separated score estimate between the top two.

The Oct. 15 text-preference snapshot placed Gemini 4 Argon (High) first at 1,532 ±9 points and Claude Opus 5.5 (High) second at 1,516 ±7. It covered 8,834,211 votes across 416 models, including 8,300 votes for Opus. This ranking compares preferences between text responses, while the agent ranking evaluates models during multi-step tasks using outcomes and feedback from real user sessions. The two results therefore describe different kinds of performance.

Anthropic introduced Claude Opus 5.5 on Sept. 22 as the first model in its Claude 5.5 family.