Anthropic launches Claude Opus 5.5 with lower cost and longer-horizon capabilities

Anthropic introduced Claude Opus 5.5 on September 22, 2026, describing it as the first model in a new Claude 5.5 family. The company positions the release as an upgrade for long-running coding, computer-use and knowledge-work tasks. Anthropic says Opus 5.5 performs better than its previous Opus 5 model on internal evaluations while requiring less compute to serve. Those results are provider-reported claims about Anthropic’s tests, not an independent ranking of every model or workflow.

Anthropic lists input tokens at $4 and output tokens at $20 per million, compared with $5 and $25 for Opus 5. Cache reads are listed at $0.20 per million tokens, down from $0.50. The company says lower token prices and fewer tokens per task produce a 40% reduction in typical workload cost, while output generation is more than 30% faster. A faster mode is also available in Claude Code and the Claude Platform at higher token rates.

Anthropic highlights agentic coding and multi-step work. Its examples include a 680,000-line code migration completed in less than a day, a web-app performance task succeeded 39 times out of 40, and a 200,000-line codebase audited and fixed in under three hours. The announcement also reports leading results on selected coding, computer-use and knowledge-work evaluations. Early testers from GitHub, Spotify and Optiver describe fewer steps, lower token use and less rework, but these are testimonials rather than controlled independent studies.

Safety is part of the launch. Anthropic says Opus 5.5 is more resistant to prompt injection than Opus 5, less likely to take hard-to-reverse actions and evaluated across longer, more realistic scenarios. The model includes safeguards for sensitive biology and cybersecurity work, with verification programs for vetted users. Organisations still need permissions, logging, human review and task-specific testing before allowing an agent to change production systems.

For developers and businesses, the release could make capable long-running agents cheaper to trial and easier to justify in production. The important questions are whether the lower cost survives real workloads, whether safeguards fit accessible tools and how often people must correct the model. Anthropic says Sonnet 5.5 and Haiku 5.5 will follow in the coming weeks, extending this efficiency and safety direction to other tiers.