Last checked: 29 October 2026 against Anthropic's pricing page.
Anthropic released Claude Opus 5.5 on September 22, 2026, and cut the price of its Opus tier for the first time. The model costs $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. It's available in the Claude API as claude-opus-5-5 and through the major cloud platforms.
The new rates
- Standard: $4 input and $20 output per million tokens (Opus 5: $5 and $25).
- Cache reads: $0.20 per million tokens, 5% of the input price and 60% below Opus 5's $0.50.
- Batch API: half price, $2 input and $10 output.
- Fast mode (up to 2.5 times faster): $8 input and $40 output.
- US-only inference: 1.1 times the standard rates.
- Context window: 1 million tokens, at no extra charge.
What it means for a real bill
The per-token cut is 20% on any workload. For an app sending 5,000 input tokens and getting 1,000 output tokens per request, 10,000 times a month:
Input
5,000 × $4.00 ÷ 1Mequals$0.02000
Output
1,000 × $20.00 ÷ 1Mequals$0.02000
Per month
$0.04000 × 10,000equals$400.00
The same traffic on Opus 5 costs $500 a month. Anthropic says typical workloads run about 40% cheaper than on Opus 5, because Opus 5.5 also uses fewer output tokens per task. That larger figure is Anthropic's own estimate; your saving depends on how long your model's answers are and how much of your input is cached.
Fable 5.1 cache reads also fell
Earlier in September, Claude Fable 5.1 launched at the same $10 input and $50 output as Fable 5, but with cache reads cut from $1.00 to $0.25 per million tokens. Workloads that resend a long system prompt or document benefit most.
Tip: Thinking tokens are billed as output. Before switching models, replay a sample of real requests at the effort level you'll use and compare the full cost per request, not just the rate card.
Cache writes, and why agents save the most
Prompt caching has two prices: writing content into the cache and reading it back. On Opus 5.5, a five-minute cache write costs $5 per million tokens and a one-hour write $8, while a cache read costs $0.20. Reads are 5% of the normal input price, a lower ratio than most Claude models.
Agent and coding workloads are dominated by cache reads, because the same growing context is resent at every step. Take a session with 10 million tokens of cache reads, 500,000 uncached input tokens and 200,000 output tokens:
- Opus 5: 10M × $0.50 + 0.5M × $5 + 0.2M × $25 = $12.50.
- Opus 5.5: 10M × $0.20 + 0.5M × $4 + 0.2M × $20 = $8.00.
That is 36% less, much closer to Anthropic's 40% figure than the 20% rate-card cut, entirely because of the cheaper cache reads.
Why your saving might be smaller
Independent testers have reported a more mixed picture: cost per task roughly level with Opus 5 on general tasks, and about 21% higher on long coding-agent runs. The difference comes from how many tokens the model spends, including thinking, which is billed as output. Opus 5.5 also defaults to medium effort, and in Claude Code a previously saved effort level does not carry over, so the settings you test with matter. The practical advice is simple: replay a sample of your real tasks at the effort level you will actually use, and compare total cost per completed task.
How it compares
- Claude Opus 5.5: $4 / $20, cache reads $0.20.
- Claude Opus 5: $5 / $25, cache reads $0.50 (still available).
- Claude Fable 5.1: $10 / $50, cache reads $0.25.
- Claude Sonnet 5: $2 / $10.
- OpenAI GPT-6 Sol: $2 / $10, released the same day; GPT-6 Astra: $10 / $50.
Opus 5.5 sits at twice the per-token price of the mid-tier models. It is worth it where it completes hard tasks in fewer attempts; for routine work, the mid-tier models usually cost less per finished job.
Questions people ask
- How much does the Claude Opus 5.5 API cost?
- $4 per million input tokens and $20 per million output tokens, with $0.20 per million for cache reads and half-price batch. Confirm the current figure on Anthropic's pricing page before you build a budget on it.
- How much does Claude Opus 5.5 cost?
- $4 per million input tokens and $20 per million output tokens, 20% below Opus 5's $5 and $25. Batch requests are half price.
- How much are Claude Opus 5.5 cache reads?
- $0.20 per million tokens, 5% of the input price.
- Is Opus 5.5 really 40% cheaper than Opus 5?
- The per-token cut is 20%. Anthropic estimates typical workloads run about 40% cheaper because Opus 5.5 uses fewer output tokens per task; your saving depends on your own requests.
- How much do Opus 5.5 cache writes cost?
- $5 per million tokens for a five-minute cache write and $8 for a one-hour write. Cache reads are $0.20 per million.
- Is Opus 5.5 always 40% cheaper than Opus 5?
- No. That is Anthropic's estimate for typical workloads. Independent tests found similar per-task cost on general tasks and higher cost on long coding runs, depending on effort settings.
Sources
Get fee-change alerts
One short email when a price our calculators use changes. No other mail, unsubscribe anytime.