Claude Opus 5.5 official: same top performance, but 40% cheaper

Anthropic has announced Claude Opus 5.5, the first model of the new 5.5 family, which in most tasks performs like Claude Fable 5.1, the company’s previous flagship model, yet costing 40% less than Opus 5.

Aimed primarily at those who use AI for complex programming, automation and office work tasks, where Anthropic talks about a clear performance leap compared to the previous generation.

Claude Opus 5.5: benchmark, speed and other news

On the programming front, Anthropic reports the case of a tester who completed the migration of a codebase of 680,000 lines in under a day, a job that previously would have required weeks of effort.

In official benchmarks, Claude Opus 5.5 achieves 66.4% on Terminal-Bench 4.0 (beating GPT-6 Astra at around 40% of its cost) and 54.4% on FrontierCode v1.1, ahead of both Fable 5.1 (50.3%) and Opus 5 (48.0%). On CursorBench 4.0 the score is 57.8%, with an 11-point lead over GPT-5.6 Sol.

The model generates responses more than 30% faster than Opus 5, while also requiring fewer tokens to complete the same tasks. Anthropic also notes a separate price list for a ‘Fast’ mode, which promises speeds of up to 2.5x.

Price: input and output cut by 20%, cache up to 60% less

The cost of Claude Opus 5.5 drops to $4 per million input tokens and $20 per million output tokens, both 20% lower than Opus 5; cache reads, however, fall by 60%, to $0.20 per million tokens, while cache writes drop by 20% to $5 per million. Overall, Anthropic cites a 40% reduction in operating costs.

Fewer attempts to bypass and greater resistance to prompt injections

On the security front, Claude Opus 5.5 attains the highest scores in Anthropic’s automated behavioral audit, conducted across nearly 2,000 scenarios, resulting in it being less prone than previous models to attempt to bypass imposed limits and more resistant to prompt-injection attacks.

Specific controls remain active for the domains of cybersecurity and biology, with the routing of the most sensitive tasks to earlier models or dedicated verification programs.

Among the initial feedback reported by Anthropic, GitHub notes that the model has solved more terminal tasks than Opus 5 in fewer than half the steps.

Disponibilite0

Claude Opus 5.5 is already available on all major platforms (AWS, Google Cloud, Microsoft Azure) and via Claude Platform with the identifier claude-opus-5-5, with zero data retention option and watermarking measures compliant with the European AI Act.

Usage limits have been expanded for the Pro, Max, Team, and Enterprise plans. Anthropic has also announced that Claude Sonnet 5.5 and Claude Haiku 5.5 will arrive in the coming weeks, with similar improvements in performance, efficiency and security.

For all the latest on the world of artificial intelligence, you can find our most recent articles in GizChina.it’s AI section.