Anthropic has released Claude Opus 5.5, the first model in its new Claude 5.5 family. The launch combines a stronger focus on long-running coding agents with lower serving costs and a broader set of safeguards for autonomous work.
Standard API pricing is $4 per million input tokens and $20 per million output tokens. That is 20% below Opus 5 on raw token rates, while Anthropic says the model’s greater efficiency brings typical workload cost down by about 40% overall.
What happened
Anthropic says Opus 5.5 generates output more than 30% faster than Opus 5. Cache reads fall to $0.20 per million tokens from $0.50, a meaningful reduction for Claude Code sessions and other agentic workflows that repeatedly reuse large contexts.
The model is aimed at codebase-wide migrations, audits and other jobs that run for hours. Anthropic highlights internal tests and early-customer examples involving very large repositories and multi-language rewrites. Those examples are company and partner reports rather than universal guarantees for every project.
Safety is a central part of the release. Frontier Design and METR evaluated the model before launch, and Anthropic reports better results on its automated behavioral audit, stronger prompt-injection resistance and fewer attempts to take hard-to-reverse actions outside assigned boundaries.
Opus 5.5 is available through Claude, Claude Code and the Claude Platform, as well as Amazon Web Services, Google Cloud and Microsoft Azure. Anthropic says Sonnet 5.5 and Haiku 5.5 are due in the coming weeks.
Why it matters
For coding agents, raw token pricing tells only part of the story. The actual bill depends on how many turns, tool calls and repeated context passes a model needs before it completes the job. A model that finishes in fewer steps can create a larger cost advantage than its headline token discount suggests.
Anthropic is also trying to pair higher capability with tighter controls for long-running agents. After recent attention on AI systems crossing evaluation boundaries, operational safety is becoming a core product feature rather than an appendix to benchmark performance.
The release adds further pressure to the model market on cost per completed task. Vendors are increasingly competing on efficiency, speed and reliability over extended workflows, not just on a single top-line benchmark score.
