Claude Opus pricing: Opus 5.5, Opus 5 and Opus 4.8
Claude Opus is Anthropic's model for long-running agentic coding and knowledge work. This page lists what each current Opus version costs, what a typical request costs, and when the step up to Fable or the step down to Sonnet is worth it. It belongs to our AI costs and pricing section.
Prices last verified against official provider pages on . How we verify prices. Written by the Chelure Enterprises Editorial Team.
Prices per million tokens
| Model | Input | Output | Cached input (read) | Cache write (5 min) | Batch API | Announced change |
|---|---|---|---|---|---|---|
| Claude Opus 5.5 | $4.00 | $20.00 | $0.20 | $5.00 | Yes (50% off) | - |
| Claude Opus 5 | $5.00 | $25.00 | $0.50 | $6.25 | Yes (50% off) | - |
| Claude Opus 4.8 | $5.00 | $25.00 | $0.50 | $6.25 | Yes (50% off) | - |
Verified November 13, 2026 against the provider's pricing page. Compare every provider in the price comparison, or model your own workload in the cost calculator.
What each model is for
| Model | Positioning (provider) | Context window | Max output |
|---|---|---|---|
| Claude Opus 5.5 | For long-running agentic coding and knowledge work; Anthropic's suggested starting point for most workloads | 1M tokens | 128K tokens |
| Claude Opus 5 | Previous Opus generation | 1M tokens | 128K tokens |
| Claude Opus 4.8 | Earlier Opus generation | 1M tokens | 128K tokens |
What a request costs
Costs per single request, from the prices above. Agent step uses cache reads at each model's cached-input rate.
| Model | Chat turn (1,500 input / 400 output tokens, no caching) | RAG answer (8,000 input / 500 output tokens, no caching) | Agent step (30,000 input / 2,000 output tokens, 80% of input read from cache) | 1,000 chat turns |
|---|---|---|---|---|
| Claude Opus 5.5 | $0.014 | $0.042 | $0.069 | $14.00 |
| Claude Opus 5 | $0.018 | $0.052 | $0.092 | $17.50 |
| Claude Opus 4.8 | $0.018 | $0.052 | $0.092 | $17.50 |
How to read the Opus price list
Opus 5.5 is both cheaper and, per Anthropic, the recommended default: it lists at $4 input and $20 output per million tokens, against $5 and $25 for Opus 5 and 4.8. Its cache reads are also cheaper, at 5 percent of the input price ($0.20) rather than the 10 percent ($0.50) charged on Opus 5 and 4.8, so a prompt that is mostly cached costs far less on 5.5. If your workload has a long fixed prefix, that gap matters more than the headline price difference.
Opus sits between Sonnet and Fable in Anthropic's lineup. Sonnet 5.5 lists at $2 and $10 and Fable 5.2 at $8 and $40, so Opus costs twice Sonnet and half Fable on both input and output. Use the Sonnet pricing page to see what the cheaper step costs, and our prompt caching guide to model the saving from reads.
Anthropic lists Opus 5.5 with adaptive thinking that is always on and a default effort of medium, so output cost includes whatever thinking the model chooses to do; raise or lower effort to trade quality against spend.
Frequently asked questions
How much does Claude Opus 5.5 cost?
Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, according to Anthropic's pricing. Cache reads cost 5 percent of the base input price ($0.20 per million), and Batch API requests are 50 percent off.
Is Opus cheaper than it used to be?
Per Anthropic's models overview, Claude Opus 4.1 was priced at $15 input and $75 output per million tokens, Opus 4.5 through Opus 5 at $5 and $25, and Opus 5.5 at $4 and $20.
What is the context window of Claude Opus 5.5?
1M tokens, with a maximum output of 128K tokens on the synchronous Messages API, per Anthropic's models overview.
Should I use Opus or Fable?
Anthropic suggests starting with Opus 5.5 for most workloads and moving to Fable 5.2 for demanding reasoning and long-horizon agentic work, or when your own evaluations on Opus 5.5 at higher effort still fall short. Fable 5.2 costs $8 input and $40 output per million tokens.
How do I lower my Opus bill?
Cache long repeated prefixes (reads are 5 percent of the input price on Opus 5.5), use the Batch API for work that can wait, and keep outputs short, since output costs five times input. See the prompt caching guide.
Sources
Primary sources opened on November 13, 2026. Prices come from our verified price table (see the methodology).