Anthropic released Claude Opus 5.5 on September 22, 2026, the first model in a new Claude 5.5 family. The company describes it as its flagship model for coding and AI agents, built with a very large memory that lets it work with long documents and codebases within a single session.
This expanded memory means the model can keep track of extensive material, such as large code repositories or lengthy documentation, without losing context partway through a task.
Anthropic says the new model is notably cheaper to run than its predecessor, particularly for tasks that involve repeated or extended use, such as long agentic sessions where the model works through many steps on its own.
The company has also reported that Opus 5.5 completes coding tasks faster and more efficiently than the previous version. In testing, it reportedly reached similar or better results while taking fewer steps and producing less unnecessary output, which Anthropic says translates into meaningful savings for developers running frequent or complex tasks. A faster mode is also available for users who prioritize speed over cost, aimed at time sensitive work.
In terms of overall capability,Opus 5.5 performs on par with its other advanced models, and highlights improvements in how clearly the model explains its reasoning and progress during long tasks. The company also indicated that smaller versions of the model, intended for different use cases and budgets, are expected to follow in the coming weeks as part of the same model family.
Opus 5.5 is available through Anthropic’s standard platforms and subscription tiers, though it is not included in the free plan. This broad availability suggests the company is positioning the model as a direct, ready to use upgrade rather than a limited preview release.
Anthropic has not shared details about how the model was built internally, and independent evaluations from outside researchers are still in early stages. As with previous releases, real world performance and cost outcomes may vary depending on the specific tasks and workloads involved, meaning some of the efficiency gains described by the company could look different once broader usage patterns emerge.
