Anthropic releases Claude Opus 5.5

Background

Claude is the family of AI models made by Anthropic. Its models come in tiers: names such as Opus, Sonnet and Haiku mark models of different size and purpose, and each new generation carries a version number. Before this release, Anthropic had already shipped Opus 5, and Claude Fable 5.1 was the model that the new release was measured against. A new model in this family matters for two reasons: what it can do, and what it costs to run.

What happened

On 22 September 2026 Anthropic released Claude Opus 5.5, the first model of the Claude 5.5 family.^[1][2] Anthropic said that on most work Opus 5.5 matched Claude Fable 5.1.^[1] Reuters also reported the launch, under the headline that Anthropic had unveiled the model.^[2]

Key details

The first detail is price. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, and the price fell to these levels with this release. Anthropic said that typical workloads cost about 40% less than on Opus 5.^[1] A token is the unit of text a model reads and writes, so a price per million tokens lets a developer or a company estimate what a large batch of work will cost.

The second detail is benchmark performance. On Terminal-Bench 4.0, a test of how well models complete tasks in a terminal, Opus 5.5 scored 66.4% at max effort. Claude Fable 5.1 scored 55.8% on the same test.^[1] Effort is a setting that controls how much work the model puts into each answer, so "max effort" means the setting was turned to its highest level for the score reported.

The third detail concerns subscribers. Subscription plans gained higher five-hour usage limits, so a person using the Claude apps could do more work inside a five-hour window before reaching a limit.^[1]

The release also set up the rest of the family. Anthropic said that Sonnet 5.5 and Haiku 5.5 would follow in the coming weeks.^[1]

Impact

For a reader, the notable point is the combination of price and score. The release lowered what it costs to run the top-tier model while reporting a higher result than Claude Fable 5.1 on a test of terminal work. These figures come from Anthropic's own announcement and should be read as the company's account of its results, not as an independent measurement. The date and the fact of the release are confirmed by both Anthropic's announcement and the Reuters report.^[1][2]