Anthropic launched Claude Opus 5.5, its first model in the Claude 5.5 family, and said GPT-6.1 SOL now runs at 40% less than Opus 5. It also said the model is more than 30% faster than Opus 5 and stronger on several benchmarks.
For Pro, Max, Team, and seat-based Enterprise users, that changes how far each five-hour limit can stretch. Anthropic also said subscription users can save a rate-limit reset for later, which gives them more control over heavier work.
Claude Opus 5.5 pricing
Anthropic set input tokens at $4 per million and output tokens at $20 per million for Claude Opus 5.5. It also priced cache reads at $0.20 per million tokens. Those numbers matter because they turn a frontier model from a fixed expense into something teams can budget around by task volume.
Anthropic said Claude Opus 5.5 performs at the level of Claude Fable 5.1 on most work. It also said the new model is comparable to Claude Mythos 5.1 in biology and cybersecurity. That gives buyers a clearer tradeoff than a generic speed claim.
Frontier Design and METR tests
Before release, Frontier Design and METR tested Claude Opus 5.5. Anthropic said the model is the strongest-performing one it has tested to date on its automated behavioral audit, and it said it succeeded 39 of 40 times when asked to cut load times across every page of a web app.
Anthropic also said an early tester described the writing by saying, "it writes the way I do." That is useful, but it is still one tester’s read on style, not proof that the model matches human judgment across all jobs.
Pro Max Team limits
Anthropic said it is increasing five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans. It also said verified cybersecurity practitioners will be able to use Claude Opus 5.5 in the coming weeks through the Cyber Verification Program in the Cyber Verification Program in the.
Vetted organizations can apply to the Life Sciences Verification Program to use Claude Opus 5.5 for biology research. Anthropic said the model is more resistant than Opus 5 to prompt injection, which matters for users who rely on it in workflows where outside text can try to steer the system off course.
