Anthropic has released Claude Opus 5.5, a new model aimed at long-running agentic coding and knowledge work.
The company says Opus 5.5 costs about 40% less per typical workload than Opus 5 while generating output more than 30% faster. API pricing is ＄4 per million input tokens and ＄20 per million output tokens, with cache reads reduced to ＄0.20 per million tokens—an important saving for coding agents that repeatedly reuse large contexts.
Opus 5.5 supports a one-million-token context window, up to 128,000 output tokens, tool use, vision, and always-on adaptive thinking. Anthropic reports substantial gains in agentic coding, including a 66.4% Terminal-Bench 4.0 score versus 52.3% for Opus 5. These benchmark and cost-per-task results are vendor-reported and should be validated against each team’s own workloads.
The model is available through Claude, the Claude API as claude-opus-5-5, Amazon Bedrock, Google Cloud, Microsoft Foundry, and GitHub Copilot. GitHub is rolling it out across Copilot’s coding agent, CLI, app, IDE integrations, web interface, and mobile clients.
Why it matters
For builders, the most meaningful improvement may be efficiency rather than a single benchmark lead. Long-running agents accumulate costs through repeated tool calls, cached context, and retries. A model that completes comparable or better work with fewer steps and cheaper cache reads could make overnight migrations, repository-wide audits, research tasks, and autonomous debugging more practical in production.
That efficiency becomes more useful when paired with operational controls. Anthropic’s Managed Agents permission gate and live control terminal show how longer-running model capability can sit behind policy checks, approvals, and real-time human supervision.