Anthropic launches Claude Opus 5.5 at $4 per million tokens

Anthropic launched Claude Opus 5.5 on Sept. 22, presenting it as a lower-cost model with performance close to Claude Fable 5.1. The model costs $4 per million input tokens and $20 per million output tokens. Anthropic also says it was 85% less likely than Opus 5 or Claude Mythos 5.1 to try to bypass testing limits. Valor and The Verge reported the release.

What Anthropic released

Anthropic says Opus 5.5 matches Fable 5.1 on most work while using fewer resources than Opus 5. Reporting from Estadão's TecMundo says the model beat Fable 5.1 and Opus 5 across the benchmarks Anthropic highlighted, but lost to OpenAI's GPT-6 Astra on AutomationBench and Terminal-Bench-Science 0.1 when all three models used high reasoning effort. Those results came from Anthropic's testing, not an independent review. (Estadão)

The cost comparison needs care. Anthropic described Opus 5.5 as 40% cheaper to run than Opus 5 for typical workloads, while Valor reported a 20% reduction in token cost. The published rates are $4 per million input tokens and $20 per million output tokens. Anthropic also lists cache rates of $0.20 per million tokens read and $5 per million tokens written. (Valor; Estadão)

For developers, Anthropic highlighted long coding jobs such as code migrations and audits. TecMundo reported the company's example of a 200,000-line codebase audit completed in under three hours with Opus 5.5, compared with about 20 hours for Opus 5. The same example used 2.5 times more tokens with the older model. Anthropic also says the new model produces clearer, more direct text and avoids excessive jargon.

Opus 5.5 is available through Claude, Amazon Web Services, Google Cloud and Microsoft Azure. It is also available to Claude subscribers without separate paid credits, and Anthropic temporarily increased five-hour usage limits for Pro, Max and Team plans. The model runs only in reasoning mode. Claude Code and Claude Platform offer an optional Fast mode that can work up to 2.5 times faster at twice the input and output rates. (Estadão)

Why it matters

The release combines a lower published price with access for existing subscribers, which could reduce the cost of extended coding sessions for developers. It also puts safety controls alongside the performance claims. Anthropic says Opus 5.5 made fewer attempts to circumvent its testing sandbox, and The Verge reported that certain cybersecurity requests are redirected to the less powerful Opus 4.8 while flagged biology requests go to Opus 5. Anthropic tested the model with outside partners including Frontier Design and METR. (The Verge)

Anthropic said Claude Sonnet 5.5 and Haiku 5.5 are planned for release in the coming weeks. (The Verge; Estadão)