Anthropic launches Claude Opus 5.5 with 40% cost cut and best safety scores to date

Anthropic has released Claude Opus 5.5, the first model in a new 5.5 generation that the company says is its most capable and most safety-aligned release yet. The announcement comes weeks after Anthropic CEO Dario Amodei published his "pace the frontier" essay calling for coordinated AI safety standards across major labs.

Opus 5.5 is a significant step beyond Opus 5 by every metric Anthropic tracks. In its own benchmarks, the model leads in agentic coding, computer use, and knowledge work — surpassing both GPT-6 Astra on AutomationBench and its own prior models on the Terminal-Bench 4.0 coding evaluation, scoring 66.4% compared to Astra's 57.9% and Opus 5's 52.3%.

Real-world demonstrations underscore the jump: one external tester completed a 680,000-line code migration in under a day — work Anthropic says would have taken an engineering team weeks. In a separate test, Opus 5.5 successfully cut load times across every page of a web application 39 out of 40 times without altering the app's behavior, a task where Opus 5 fell short.

Pricing drops substantially alongside the performance gains. Input and output tokens are priced at $4 and $20 per million — 20% lower than Opus 5. Cache reads, which dominate costs in agentic and coding workloads, fall to $0.20 per million, a 60% cut. Anthropic also says Opus 5.5 generates output more than 30% faster than its predecessor, and is raising usage limits for Pro, Max, Team, and Enterprise subscribers.

Safety is the most prominent theme in Anthropic's announcement, which is notable given the "pacing" debate currently running through the industry. Opus 5.5 achieved the highest scores Anthropic has recorded on its automated behavioral audit — a suite that tests the model across thousands of simulated scenarios including long-horizon tasks and scenarios modeled on real incidents. It is less likely than Opus 5 to take hard-to-reverse actions, more resistant to prompt injection, and better at staying within assigned boundaries. External evaluators Frontier Design and METR tested the model before release.

Because Opus 5.5's capabilities in biology and cybersecurity are comparable to Claude Fable 5.1, Anthropic is applying the same safeguards it uses for that model. Vetted organizations can apply to the Life Sciences Verification Program for biology research access; a Cyber Verification Program expansion for security practitioners is coming in the weeks ahead.

Claude Sonnet 5.5 and Haiku 5.5 are expected within weeks, completing the 5.5 tier across the model family, according to Anthropic.