Claude Opus 5.5 reduces cost and accelerates generation

Claude Opus 5.5 succeeds Opus 5 with better performance in agentic programming, computer use, and knowledge work. Anthropic claims a 40% lower cost per task, over 30% faster generation, and is strengthening external evaluations prior to release.

Faster and cheaper

Better performance than Opus 5 with fewer resources. That is the main change behind Claude Opus 5.5, Anthropic’s new high-end model, with improvements across agentic coding, computer use, and knowledge work alongside lower pricing.

API pricing drops to $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. The 20% price cut only accounts for part of the difference. According to figures provided by Anthropic and reported by Reuters, Opus 5.5 also uses fewer resources to complete a task. At default settings, Anthropic estimates that typical workloads cost about 40% less than with Opus 5. Output generation is also more than 30% faster.

The distinction matters. Opus 5 was priced at $5 per million input tokens and $25 per million output tokens. Anthropic’s claimed 40% reduction for Opus 5.5 refers to the cost of completing typical tasks in its tests, rather than a 40% reduction in the price of each token. Default effort takes center stage

Anthropic is putting more emphasis on the relationship between performance and the amount of compute allocated to a task. At its default effort setting, Opus 5.5 is said to reach frontier-level results while using fewer resources. The company also says it outperforms Opus 5 and Fable 5.1 across most of the evaluations it reports, with particularly strong results on agentic workloads.

The changes also extend to writing behavior. In a series of posts from Claude on 𝕏, Anthropic says it addressed recurring feedback about Opus 5. The new model is designed to surface important information earlier, follow user-defined writing rules more consistently, and communicate more naturally over long sessions.

For now, most of these performance and efficiency comparisons come from Anthropic’s own evaluations. They should therefore be separated from independent comparisons conducted under identical conditions. The first model since Anthropic called for pacing the frontier

Opus 5.5 also arrives in a different context for Anthropic. It is the company’s first model release since its public call to slow the pace of frontier development and leave more room for evaluation and safety measures.

Before release, the model was tested by external evaluators including METR and Frontier Design. Anthropic says Opus 5.5 achieved its strongest result to date on its most comprehensive alignment evaluation. Reuters also reports that the model was 85% less likely than Opus 5 and Mythos 5.1 to attempt to circumvent operational boundaries during testing. That figure comes from Anthropic’s evaluation framework and should not be read as a general measure of model safety.

Targeted safeguards remain in place for sensitive capabilities. Higher-risk cybersecurity requests can be routed to Opus 4.8, while certain biology-related requests can similarly be redirected to Opus 5. Higher limits for Claude subscribers

The release also comes with higher usage limits for Pro, Max, and Team subscriptions. Anthropic is increasing the limits applied to five-hour usage windows and giving subscribers a rate-limit reset that can be saved and used when needed.

The move extends an earlier increase in capacity. Anthropic had already doubled some Claude Code limits for Pro, Max, and Team accounts following an expansion of its available compute infrastructure.