Anthropic Announces Claude Opus 5.5 With 40% Lower Typical Workload Costs

Anthropic has announced Claude Opus 5.5, the first model in its Claude 5.5 family. According to Anthropic, Opus 5.5 reaches Claude Fable 5.1-level performance across most work while costing 40% less than Opus 5 on typical workloads.

For visual creators and technical teams, the main change is efficiency on complex, multi-step work. Anthropic says an early tester completed a 680,000-line code migration in less than a day, a project that would previously have required an engineering team several weeks. In another example, Opus 5.5 shortened the loading time of a web application successfully in 39 of 40 attempts, while Opus 5 made only minor improvements and changed the application’s behavior. Anthropic also says an early tester asked several Claude models to build a game from a single prompt, with Opus 5.5 receiving the highest score for visual presentation and detail.

The published pricing is $4 per 1M input tokens and $20 per 1M output tokens, compared with $5 and $25 for Opus 5. Cache reads cost $0.20 per 1M tokens, down from $0.50, while cache writes cost $5, down from $6.25. Anthropic reports that output generation is more than 30% faster. A fast mode is available in Claude Code and Claude Platform at $8 per 1M input tokens and $40 per 1M output tokens, with speeds of up to 2.5 times faster.

Anthropic’s published benchmark table reports scores of 66.4% on Terminal-Bench 4.0, 57.8% on CursorBench 4.0, 81.8% partial on OSWorld 2.0, and 89.0% with tools on Chartography. On GDPval-AA v2.1, which covers 44 professions, Opus 5.5 scored 1846 Elo. Anthropic cautions that benchmark gaps may no longer reliably represent real-world differences at these capability levels. Its benchmark notes also identify results reported by OpenAI or Zapier, standard errors, effort settings, and cases where safety measures routed tasks to other models.

Safety is a central part of the release. Anthropic says Opus 5.5 was evaluated before launch by Frontier Design and METR and achieved its strongest result so far in an automated behavioral audit covering nearly 2,000 simulated scenarios. The company reports that attempts to cross isolation boundaries were about 85% less frequent than with Opus 5 or Claude Mythos 5.1, while acknowledging that evaluations have limitations.

For creators considering AI agents for production workflows, the practical implication is a potentially lower cost for long sessions involving code, research, visual prototyping, or structured deliverables. That is AiPix analysis based on Anthropic’s claims, not independent testing. Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will arrive in the coming weeks, while Opus 5.5 is already available across all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure.