AI-generated illustration
Anthropic has officially launched Claude Opus 5.5, representing the first model in its new 5.5 generation of systems. The release performs at the level of Claude Fable 5.1 across most tasks while cutting operational compute expenses by 40 percent compared to Opus 5. Evaluators from organizations including Frontier Design and METR conducted external assessments prior to the public deployment.
Claude Opus 5.5 Operational Cost and Speed Improvements
The updated model reduces token pricing significantly compared to its predecessor. Input tokens cost $4 per million, while output tokens are priced at $20 per million, reflecting a 20 percent decrease. Furthermore, cache reads cost $0.20 per million tokens, delivering a 60 percent drop in expenses for agentic development workflows in software applications.
In addition to cost savings, output generation speeds increase by more than 30 percent over Opus 5. The provider also introduced an optional fast mode across its platform, operating at up to 2.5 times standard speed for $8 per million input tokens and $40 per million output tokens.
Benchmark Results Across Technical Evaluations
Technical evaluations show that Claude Opus 5.5 attained a score of 66.4 percent on Terminal-Bench 4.0, surpassing GPT-6 Astra at 57.9 percent. On the FrontierCode benchmark, the system registered 54.4 percent accuracy, while reaching 57.8 percent on CursorBench 4.0. In general knowledge testing via GDPval-AA v2.1 across 44 occupations, the model achieved 1846 Elo.
“Developers want agents that can take on real software work and finish it. In our testing across GitHub Copilot CLI and VS Code, Claude Opus 5.5 used among the fewest tokens and steps we measured. In VS Code, it solved more terminal tasks than Opus 5 in less than half the steps.”
Mario Rodriguez, Chief Product Officer at GitHub
Security Controls and Access Programs
To mitigate risks, the deployment integrates an action-screening classifier, an open-source sandbox for enterprise audits, and automated vulnerability detection. In external evaluations conducted by Gray Swan, the model matched Fable 5.1 for the lowest prompt injection success rate recorded. Anthropic is also expanding vetting programs in cybersecurity and life sciences research to maintain operational safeguards.
Future Expansion of the Model Family
Developers utilizing the model can access increased five-hour rate limits across commercial subscription tiers. Anthropic confirmed that Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in upcoming weeks, extending these technical updates across its broader artificial intelligence portfolio.
