Claude Opus 5 and 5.5: Pricing, Release Facts, and How to Compare

Anthropic released Claude Opus 5 on July 24, 2026. On September 22, Anthropic announced Opus 5.5, so Opus 5 is no longer the latest Opus release. This article updates the original launch-day comparison with current provider information and distinguishes per-token prices from Anthropic's estimates of cost across typical workloads.

Anthropic's Opus 5 release page lists the API model ID as claude-opus-5 and a price of $5 per million input tokens and $25 per million output tokens. It also describes a Fast mode priced at twice the standard token rates. Anthropic's release announcement presents Opus 5 as an improvement for coding, knowledge work, and long-running agents; its benchmark and customer stories are vendor-reported evidence, not independent tests by AcceleratedLogic.

Anthropic's Opus 5.5 announcement lists $4 per million input tokens, $20 per million output tokens, and $0.20 per million cache-read tokens. It says Opus 5.5 costs about 40% less to run than Opus 5 for typical token-billed workloads. The lower per-token price is one part of that estimate; Anthropic also attributes savings to using fewer tokens per task. The overall saving will depend on a workload's cache-hit rate, effort setting, and output length.

The two model releases also have different model IDs and evaluation snapshots. When comparing them, do not reuse Opus 5's benchmark or pricing figures for Opus 5.5. Consult the Opus product page and the API model deprecation table to verify current availability before making an integration change.

A per-token price does not answer which model is cheaper for a particular job. Use the same tasks, prompts, tools, and success criteria for each candidate. Record input and output usage, cache reads, latency, retries, and human corrections, then calculate cost per accepted result. Include a representative mix of short requests and the longer jobs for which caching or higher effort settings may matter.

For example, a model with a lower cache-read price may save more on a repeated long context than on a short exchange that is mostly new input. A faster mode may be justified for an interactive workflow but wasteful for a batch that does not have a deadline. Measure those trade-offs on your traffic rather than assuming the release's typical-workload estimate applies to your application.

Opus 5 was introduced at $5/$25 per million input/output tokens, and Opus 5.5 is the newer Opus release with lower listed rates and an Anthropic-estimated reduction in typical workload cost. These are useful starting facts, not a universal model ranking. Recheck Anthropic's live pricing and availability pages and run a task-specific comparison before changing a production default.