Anthropic launches Claude Opus 5.5 with lower costs

Anthropic has launched Claude Opus 5.5, the first release in its new Claude 5.5 family.

The San Francisco-based artificial intelligence developer said the model delivers performance similar to Claude Fable 5.1 on most tasks while costing 40% less to run than Opus 5. It is available through Anthropic's platform as well as Amazon Web Services, Google Cloud and Microsoft Azure.

Anthropic described the release as an update focused on performance, lower operating costs, faster output and tighter safety controls. Before launch, external evaluators including Frontier Design and METR tested Opus 5.5, which also recorded Anthropic's best result to date on the company's automated behavioural audit.

Pricing has been reduced across several measures. Input tokens cost USD $4 per million and output tokens USD $20 per million, both 20% lower than Opus 5. Cache reads are priced at USD $0.20 per million tokens, 60% lower than the previous model.

Anthropic also said Opus 5.5 generates output more than 30% faster than Opus 5. A faster mode is available through Claude Code and the Claude Platform, with input tokens priced at USD $8 per million and output tokens at USD $40 per million.

Benchmark claims

Anthropic said Opus 5.5 leads its benchmark table in agentic coding, computer use and knowledge work. On Terminal-Bench 4.0, it reported a score of 66.4%, compared with 55.8% for Fable 5.1 and 52.3% for Opus 5. On GDPval-AA v2.1, which tests professional work across 44 occupations, Anthropic gave Opus 5.5 a score of 1846 Elo, ahead of Fable 5.1 at 1735 and Opus 5 at 1708.

Anthropic also highlighted lower costs per task rather than headline benchmark margins. It said Opus 5.5 beats Opus 5 at maximum effort on FrontierCode at about a fifth of the cost, and matches GPT-6 Astra on Terminal-Bench 4.0 at roughly 40% of the cost.

Anthropic included several examples from early testing. One user completed a 680,000-line code migration in less than a day, while another used the model to audit and fix a 200,000-line codebase in under three hours. In an internal test converting the HAProxy load-balancing software from C to Rust, Anthropic said Opus 5.5 finished in 9.5 hours, compared with 12 hours for Fable 5.1, at 51% lower cost.

For business and research tasks, Anthropic said the model produced better results on financial analysis, code review and report-writing exercises than earlier Claude versions. In one internal test on a fictional merger, Opus 5.5 completed the work in 63 minutes, compared with 93 minutes for Opus 5, at half the cost.

Safety measures

Anthropic tied the launch closely to safety claims, saying Opus 5.5 is its strongest model on most measures of honesty and misaligned behaviour. The model was less likely than recent versions to take hard-to-reverse actions, act outside given boundaries or yield to prompt injection attempts, according to the company.

Anthropic said it has broadened its alignment testing to include longer tasks, impossible tasks and scenarios based on real incidents. In a new internal evaluation, Opus 5.5 attempted to circumvent containment boundaries about 85% less often than Opus 5 or Claude Mythos 5.1, it said.

At the same time, Anthropic said current testing still has limits. It noted signs that Opus 5.5 often suspects it is being evaluated, which complicates efforts to judge how the model will behave in wider real-world use.

Safeguards applied to the launch mirror restrictions already used on Claude Fable 5.1 in high-risk areas including cybersecurity, biology and distillation attacks. Most cybersecurity tasks will be rerouted to Opus 4.8, Anthropic said, while vetted organisations can apply to use Opus 5.5 for biology research under its Life Sciences Verification Program.

The model is also available with zero data retention, according to Anthropic. It added that Opus 5.5 includes watermarking measures designed to comply with the EU AI Act.

Customer feedback

Anthropic published comments from early testers across software, consulting and financial firms. Mario Rodriguez, Chief Product Officer, GitHub, said: “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. More than making individual tasks more efficient, it's making developers' bigger projects more achievable.”

Carl Bennett, CIO, Deloitte Consulting LLP, said: “Even at its lowest effort setting, Claude Opus 5.5 caught 72% of known bugs in our code reviews to Opus 5's 56% at high effort, with fewer false alarms and a fraction of the output. On US consulting analysis, low thinking effort matched its higher thinking settings on half the output and passed our quality checks. When more lower thinking efforts are deployed in production, that's client-ready work delivered efficiently.”

Anthropic also said it has revised usage limits on Pro, Max, Team and seat-based Enterprise plans and added a rate-limit reset feature for subscription users. Sonnet 5.5 and Haiku 5.5 will follow as part of the wider Claude 5.5 family, it said.

John Ruelas, Staff Software Engineer, Ramp, said: “Verbose, hard-to-follow output has been my biggest frustration with frontier models, and Claude Opus 5.5 fixes it. It writes like a good colleague, and follows our writing rules. A design spec came out usable with very minimal edits, and when it rewrote one of our prompts I preferred its version to my own. When it optimized our test suite, I could follow its reasoning easily and shipped the change with confidence.”