Key takeaways
- Claude Opus 5.5 lists at $4.00 per million input tokens and $20.00 per million output tokens, 20% below Opus 5.
- Cached input reads cost $0.20 per million tokens and cache writes cost $5.00 per million, so repeated system prompts get dramatically cheaper.
- A faster output mode is available at $8.00 input and $40.00 output, double the standard rate.
- For a typical 2,000 in / 500 out request, Opus 5.5 costs about $0.018, versus $0.0225 on Opus 5 and $0.045 on Fable 5.1.
- Anthropic reports output that is more than 30% faster and typical jobs costing about 40% less. These are vendor figures, so test on your own workload before switching.
Anthropic launched Claude Opus 5.5 on 22 September 2026 as the first model in its new 5.5 family. The headline for anyone paying the bill is simple: it is cheaper than Opus 5 per token, and Anthropic positions it close to Fable 5.1 on most everyday work. This guide sticks to what you can verify, the published rate card and the billing mechanics, and turns it into numbers you can multiply by your own traffic.
If you want to run the arithmetic while you read, open the Claude cost calculator and the pricing database in another tab. Both use the same per-million-token rates quoted here.
The Claude Opus 5.5 rate card
Anthropic bills Opus 5.5 per million tokens, with separate prices for fresh input, cached input, cache writes and output. A token is roughly three quarters of an English word, so a million tokens is around 750,000 words.
| Billing item | Standard | Fast mode |
|---|---|---|
| Input tokens | $4.00 | $8.00 |
| Output tokens | $20.00 | $40.00 |
| Cached input (read) | $0.20 | Not published |
| Cache write | $5.00 | Not published |
Output costs five times as much as input. That ratio matters more than the headline number: a workload that writes long answers is dominated by output cost, while a workload that reads large documents and returns short answers is dominated by input cost.
How it compares with Opus 5 and Fable 5.1
| Model | Input | Output | Relative to Opus 5.5 |
|---|---|---|---|
| Claude Opus 5.5 | $4.00 | $20.00 | Baseline |
| Claude Opus 5 | $5.00 | $25.00 | 25% more expensive |
| Claude Fable 5.1 | $10.00 | $50.00 | 2.5x more expensive |
| Opus 5.5 fast mode | $8.00 | $40.00 | 2x more expensive |
Put another way, Opus 5.5 is 20% cheaper than Opus 5 and 60% cheaper than Fable 5.1 per token. Whether that saving survives in practice depends on how many tokens each model spends to finish the same task, which we cover in our Opus 5.5 vs Fable 5.1 comparison.
What a single request costs
The cleanest way to compare models is cost per request. Take a common shape: a 2,000 token prompt (instructions plus some context) and a 500 token answer.
- Opus 5.5: 2,000 x $4 / 1M = $0.008 input, plus 500 x $20 / 1M = $0.010 output, for $0.018 per request.
- Opus 5: $0.010 input plus $0.0125 output, for $0.0225 per request.
- Fable 5.1: $0.020 input plus $0.025 output, for $0.045 per request.
- Opus 5.5 fast mode: $0.016 input plus $0.020 output, for $0.036 per request.
At 100,000 requests a month that becomes roughly $1,800 on Opus 5.5, $2,250 on Opus 5, $3,600 in fast mode and $4,500 on Fable 5.1.
Prompt caching is the biggest lever
Most production prompts repeat a large fixed block: a system prompt, a style guide, tool definitions or a reference document. Caching lets you pay $5 per million once to write that block, then $0.20 per million each time it is read back, which is 95% below the fresh input rate.
Using the same 2,000 in / 500 out request, suppose 1,500 of the input tokens are a stable cached prefix. The input cost falls from $0.008 to $0.0023 (1,500 cached tokens at $0.20 plus 500 fresh tokens at $4). The request drops to about $0.0123, or around $1,230 a month at 100,000 requests. Output is untouched, so caching helps most on read-heavy work.
Caching only pays when the prefix is reused before it expires. A prompt that changes on every call gains nothing and adds the write cost.
When fast mode is worth double
Fast mode charges twice the standard rate for quicker output. It is worth it where a person is waiting: live chat, in-editor coding help, or voice interfaces where latency drives abandonment. For batch jobs, overnight pipelines and background agents, the standard tier does the same work for half the price.
Worked monthly budgets
| Workload | Tokens per request | Requests / month | Monthly cost |
|---|---|---|---|
| Support chatbot | 1,500 in / 300 out | 50,000 | $600 |
| Document summaries | 12,000 in / 800 out | 10,000 | $640 |
| Coding agent step | 40,000 in / 4,000 out | 5,000 | $1,200 |
| Content drafting | 1,000 in / 2,000 out | 5,000 | $220 |
Replace these with your own numbers in the token calculator if you are unsure how many tokens your prompts use, then plug the totals into the AI ROI calculator to see whether the spend pays for itself.
What Anthropic claims about performance
Anthropic says Opus 5.5 produces output more than 30% faster than Opus 5 and that typical jobs cost around 40% less once fewer retries and shorter outputs are counted. In one code migration example it published, Opus 5.5 finished in 9.5 hours against 12 hours for Fable 5.1 at 51% lower cost. These are vendor-reported results on vendor-chosen tasks. They are a reason to run your own test, not a guarantee.
Who should switch
- Teams on Opus 5: switching is a straightforward 20% per-token saving. Run a short regression test on your prompts, then move.
- Teams on Fable 5.1 for routine work: test Opus 5.5 on a sample. If quality holds, the per-token rate is 60% lower.
- Teams on Sonnet or Haiku tiers: Opus 5.5 is still more expensive. Only upgrade traffic where the cheaper model visibly fails.
To see how Opus 5.5 stacks up against models from other providers, use the model comparison tool.
Frequently asked questions
- How much does Claude Opus 5.5 cost?
- $4.00 per million input tokens and $20.00 per million output tokens on the standard tier. Cached input reads cost $0.20 per million and cache writes $5.00 per million.
- Is Claude Opus 5.5 cheaper than Opus 5?
- Yes. Its per-token rates are 20% lower than Opus 5, which lists at $5.00 input and $25.00 output per million tokens.
- What is Opus 5.5 fast mode?
- A faster output option priced at $8.00 input and $40.00 output per million tokens, double the standard rate. It suits latency-sensitive, user-facing work.
- Are Anthropic's speed and cost claims independently verified?
- Not yet. The 30% speed gain and 40% lower job cost are Anthropic's own figures. Test on a sample of your real prompts before committing.
About the author
AIOPLY Pricing Desk
Independent AI cost research, verified against provider documentation
Every figure in this article was checked against the provider's own pricing page before publication. Where AI assistance is used to draft a routine price report, a human editor verifies the numbers and signs it off.
Our editorial policyQuality standards
Prices were checked against provider documentation on Sep 19, 2026. Rates change without notice, so confirm current figures with the provider before committing a budget. We publish list prices only and take no payment for placement.
Report a correctionSources and further reading
More from the blog
Comparisons
Claude Opus 5.5 vs Fable 5.1 | Cost per Task and Which to UseOpus 5.5 is 60% cheaper per token than Fable 5.1. We compare pricing, cost per task, speed claims and where each model still earns its place in your stack.
Model releases
GPT-6 Astra Pricing | $10/$50 per M Tokens ComparedGPT-6 Astra costs $10 input and $50 output per million tokens, 2.5x GPT-5.6 Sol and above Claude. Real cost per task, long-context billing, and benchmarks.
Pricing updates
Grok API Pricing Explained: What xAI Models Actually Cost to RunA working breakdown of Grok API pricing per million tokens, including cached input, long-context billing, tool calls, and how to forecast a monthly bill.