Anthropic announced Claude Opus 5.5 on September 22, 2026, the first model in its Claude 5.5 family. The company said it performs at a level comparable to Claude Fable 5.1 across many tasks. For typical workloads under standard settings, execution costs are 40% lower than with Opus 5. The original summary notes that the article headline and publication date are September 20, while its text gives September 22 as the announcement date.

According to Anthropic’s official product page, API prices are $4 per million input tokens and $20 per million output tokens, each 20% below Opus 5. Cached reads cost $0.20 per million tokens, a 60% reduction. The 40% execution-cost figure reflects both lower per-token prices and fewer tokens needed to complete tasks; Anthropic says costs will not fall by 40% for every workload.

In Anthropic’s comparisons, Opus 5.5 scored 66.4% on Terminal-Bench 4.0, which measures command-line work, versus 52.3% for Opus 5. It scored 57.8% on the coding evaluation CursorBench 4.0, compared with Opus 5’s 46.6%. Those scores used the “xhigh” effort setting for Terminal-Bench and “max” for CursorBench, which allocate computation to model reasoning; both differ from the standard “medium” setting. Another Claude model handled some tasks where safety protections activated. Anthropic cautioned that benchmark gaps may not directly reflect differences in practical work.

In its official announcement, Anthropic said output speed is more than 30% higher than Opus 5. It also said the model was improved to put important information first in its writing and follow users’ specified writing rules. METR, Frontier Design and other outside organizations evaluated the model before release. Opus 5.5 is available on Pro, Max, Team and Enterprise plans, and for developers through Claude Platform, Amazon Web Services, Google Cloud and Microsoft Foundry. It includes safety measures similar to Fable 5.1 in areas such as cybersecurity and biology.

Official technical documentation lists a 1-million-token context length and a usual maximum output of 128,000 tokens. The model accepts text and images as input and produces text. Its Claude API model ID is “claude-opus-5-5.” Anthropic says migrations from existing systems require attention to API specification changes: Opus 5.5 cannot disable its thinking feature, and a setting that forces the use of a specific tool returns an error. The company said Claude Sonnet 5.5 and Claude Haiku 5.5 are also due within the next few weeks.

Anthropic announced Claude Opus 5.5 on September 22, 2026, the first model in its Claude 5.5 family. The company said it performs at a level comparable to Claude Fable 5.1 across many tasks. For typical workloads under standard settings, execution costs are 40% lower than with Opus 5. The original summary notes that the article headline and publication date are September 20, while its text gives September 22 as the announcement date.

According to Anthropic’s official product page, API prices are $4 per million input tokens and $20 per million output tokens, each 20% below Opus 5. Cached reads cost $0.20 per million tokens, a 60% reduction. The 40% execution-cost figure reflects both lower per-token prices and fewer tokens needed to complete tasks; Anthropic says costs will not fall by 40% for every workload.

In Anthropic’s comparisons, Opus 5.5 scored 66.4% on Terminal-Bench 4.0, which measures command-line work, versus 52.3% for Opus 5. It scored 57.8% on the coding evaluation CursorBench 4.0, compared with Opus 5’s 46.6%. Those scores used the “xhigh” effort setting for Terminal-Bench and “max” for CursorBench, which allocate computation to model reasoning; both differ from the standard “medium” setting. Another Claude model handled some tasks where safety protections activated. Anthropic cautioned that benchmark gaps may not directly reflect differences in practical work.

In its official announcement, Anthropic said output speed is more than 30% higher than Opus 5. It also said the model was improved to put important information first in its writing and follow users’ specified writing rules. METR, Frontier Design and other outside organizations evaluated the model before release. Opus 5.5 is available on Pro, Max, Team and Enterprise plans, and for developers through Claude Platform, Amazon Web Services, Google Cloud and Microsoft Foundry. It includes safety measures similar to Fable 5.1 in areas such as cybersecurity and biology.

Official technical documentation lists a 1-million-token context length and a usual maximum output of 128,000 tokens. The model accepts text and images as input and produces text. Its Claude API model ID is “claude-opus-5-5.” Anthropic says migrations from existing systems require attention to API specification changes: Opus 5.5 cannot disable its thinking feature, and a setting that forces the use of a specific tool returns an error. The company said Claude Sonnet 5.5 and Claude Haiku 5.5 are also due within the next few weeks.