Anthropic released Claude Opus 5.5 on September 22. It says the model works at about the level of Claude Fable 5.1 on most work and beats it on agentic coding, while the API price drops 20% from Opus 5 to $4 per million input tokens and $20 per million output tokens. We graded it the next day at medium effort, as the successor to Claude Opus 5.
What the results show
- It is the best agent on the bench. Claude Opus 5.5 takes the top of the board, ahead of Opus 5 and Claude Fable 5.1. It fixes more real issues than either, keeps Opus 5's lead on one-line bug reports, and now writes plans as good as Fable's.
- It is light on the plan. On the same tasks it used about two thirds of Opus 5's five-hour window and about a seventh of Fable's. At the same monthly price, a run of our suites takes about 6% of Opus 5.5's weekly limit, against about a sixth for GPT-5.6 Sol and nearly a third for GPT-6 Astra.
We suspect we were already getting this model a few days early under the Opus 5 name. Opus 5 jumped to the top of the board in the weekly run the weekend before the release while answering about twice as fast. That is only our guess, but it fits. Our take: Opus 5.5 makes the $20 Claude plan the one to pick. It competes with Astra and Sol on quality, uses less of its limit, and leaves little reason to reach for Fable. GPT-6 Sol, GPT-6 Luna and Grok 4.7 came out at the same time and are still being graded. We will post their results as they come in.
Grades are comparable within the same suite version. See the full Claude Opus 5.5 scorecard or follow the family over time on Trends.