How much does the Claude Opus API cost per token?
Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, and every older Opus still on the API costs $5 and $25. We read each of the rows below from Anthropic's pricing docs on 14 November 2026. All of the prices are in US dollars per million tokens, on the default global route. Opus 4.1 and Opus 4 sit in the last row because their old price still turns up in search results, and you may see it quoted as if it were current. It isn't.

| Opus version | Input | 5-minute cache write | 1-hour cache write | Cache hit | Output | Batch input | Batch output | Context |
|---|---|---|---|---|---|---|---|---|
| Opus 5.5 | $4 | $5 | $8 | $0.20 | $20 | $2 | $10 | 1M |
| Opus 5 | $5 | $6.25 | $10 | $0.50 | $25 | $2.50 | $12.50 | 1M |
| Opus 4.8 | $5 | $6.25 | $10 | $0.50 | $25 | $2.50 | $12.50 | 1M |
| Opus 4.7 | $5 | $6.25 | $10 | $0.50 | $25 | $2.50 | $12.50 | 1M |
| Opus 4.6 | $5 | $6.25 | $10 | $0.50 | $25 | $2.50 | $12.50 | 1M |
| Opus 4.5 | $5 | $6.25 | $10 | $0.50 | $25 | $2.50 | $12.50 | 200K |
| Opus 4.1 and Opus 4, retired | $15 | $18.75 | $30 | $1.50 | $75 | $7.50 | $37.50 | 200K |
Output costs five times as much as input on every row. That's the one rule to remember. Output costs more. Opus 5.5 is the cheapest Opus in the table. It's also the only one where a cache hit costs 5% of the input price instead of 10%. Anthropic's own advice on its models page is to "start with Claude Opus 5.5 for most workloads."
The context windows come from Anthropic's models overview: 1M tokens on Opus 4.6 and later, 200K on Opus 4.5. Opus has no long-prompt surcharge, so a big prompt costs just what you'd expect it to. Anthropic's docs say "A 900k-token request is billed at the same per-token rate as a 9k-token request."
What do 1,000 Claude Opus messages cost?
1,000 short Opus 5.5 messages cost $14, against $17.50 on Opus 5 and $52.50 on the retired Opus 4.1. A short message here is 1,000 tokens in and 500 out, priced at the list rates above. Of all the levers you can pull, fast mode and Batch move the number the most:
Here's the math for Opus 5.5, and it's the same for any model. 1,000 input tokens at $4 per million is $0.004. 500 output tokens at $20 per million is $0.010. That's $0.014 a message. Longer jobs scale the same way.
| Per 1,000 messages | Short chat, 1,000 in and 500 out | Long document, 20,000 in and 1,000 out | Same document on Batch |
|---|---|---|---|
| Opus 5.5 | $14 | $100 | $50 |
| Opus 5, 4.8, 4.7, 4.6, 4.5 | $17.50 | $125 | $62.50 |
| Opus 4.1 and 4, retired | $52.50 | $375 | $187.50 |
A chat also sends its whole history again on each turn. Turn ten of a thread carries the nine before it as input, so a real conversation costs more than what you see in the short column. Our Claude API pricing page puts the same short message next to Sonnet, Haiku and Fable.
Is Opus 5.5 cheaper than Opus 4.6 for the same work?
Not at Anthropic's own token figure. Opus 5.5 costs about 4% more than Opus 4.6 for the same text, even though its list price is 20% lower. Anthropic's docs explain: "This tokenizer produces approximately 30% more tokens for the same text." The new tokenizer starts with Claude 4.7, so Opus 4.6 and Opus 4.5 use the old one. So we took a message that measures 1,000 tokens in and 500 out on the old tokenizer, added 30% for the new one, and priced 1,000 of them:
| Same 1,000 messages | Tokens counted | Cost |
|---|---|---|
| Opus 4.6, old tokenizer | 1,000 in, 500 out | $17.50 |
| Opus 5.5, new tokenizer | 1,300 in, 650 out | $18.20 |
| Opus 4.8, new tokenizer | 1,300 in, 650 out | $22.75 |
| Opus 4.1, old tokenizer, retired | 1,000 in, 500 out | $52.50 |
Anthropic says the exact increase "depends on the content and workload shape," so we haven't claimed that 30% holds for your prompts. Count them yourself, with the token counting endpoint before you budget a move. Against Opus 4.7, 4.8 or 5, which already use the new tokenizer, Opus 5.5 is a straight 20% cut. This table is about cost. It doesn't say which model does the job better.
Which Claude Opus versions can you still call, and until when?
Six Opus versions are active on the Claude API and two are gone. Opus 4.1 was retired on 5 August 2026 and Opus 4 on 15 June 2026, and Anthropic names Opus 4.8 as the replacement for both of them. Opus 4.5 is next, and Anthropic lists its tentative retirement as "not sooner than November 24, 2026." We read these dates on Anthropic's model deprecations page on 14 November 2026.
| Opus version | API model name | Status | Tentative retirement |
|---|---|---|---|
| Opus 5.5 | claude-opus-5-5 | Active | Not sooner than 22 September 2027 |
| Opus 5 | claude-opus-5 | Active | Not sooner than 24 July 2027 |
| Opus 4.8 | claude-opus-4-8 | Active | Not sooner than 28 May 2027 |
| Opus 4.7 | claude-opus-4-7 | Active | Not sooner than 16 April 2027 |
| Opus 4.6 | claude-opus-4-6 | Active | Not sooner than 5 February 2027 |
| Opus 4.5 | claude-opus-4-5-20251101 | Active | Not sooner than 24 November 2026 |
| Opus 4.1 | claude-opus-4-1-20250805 | Retired | 5 August 2026 |
| Opus 4 | claude-opus-4-20250514 | Retired | 15 June 2026 |
Anthropic promises at least 60 days' notice before it retires a public model. That's a comfort, but not a plan. Moving off of Opus 4.5 trims the list price by $1 in and $5 out, and it swaps a 200K context window for one of 1M. Opus 4.1 and Opus 4 can still run on some of the partner clouds, where Amazon Bedrock and Google Cloud set their own prices and dates.
How much do caching, Batch and fast mode change an Opus bill?
Caching and Batch together cut a repeated Opus 5.5 job from $32 to $6.50, while fast mode doubles the price. A cache hit costs $0.20 per million tokens on Opus 5.5 and $0.50 on the older versions. Batch halves both the input and the output. Anthropic says the cache multipliers "stack with other pricing modifiers, including the Batch API discount and data residency." Here's a job that classifies 1,000 support tickets with the same 5,000-token instructions, 500 tokens of ticket and 500 tokens out:
| 1,000 tickets | Opus 5.5 | Opus 4.8 |
|---|---|---|
| No caching | $32.00 | $40.00 |
| Cache hits on the instructions | $13.00 | $17.50 |
| Cache hits on the Batch API | $6.50 | $8.75 |
Writing the cache costs extra, but only once. The first request that fills it will pay 1.25x the input price for a 5-minute cache, or $0.025 for those 5,000 tokens on Opus 5.5. A 1-hour cache costs 2x to write, which pays off after the second read of it. Batch results aren't instant. Keep it for work that can wait, such as nightly classification, bulk summaries or an eval run that nobody is watching live. That's all it's for.
Fast mode gives up to 2.5x faster output at a premium. It's $8 in and $40 out on Opus 5.5, and $10 in and $50 out on Opus 5 and Opus 4.8. You can't combine it with Batch, and Anthropic says that it is deprecated on Opus 4.8. US-only inference adds 1.1x to every line on Claude 4.6 and later, so 1,000 short Opus 5.5 messages cost $15.40 there. Anthropic gives new accounts free credits "to test the API," then bills actual use.
Is the API the cheapest way to use Opus?
The API is the cheapest way to use Opus if you're building an app or sending a few hundred messages. It's the wrong tool if you just want to chat with Opus, because you'd have to pay for tokens, write the code and keep a key safe. Most people don't need that. Our Claude API vs subscription page works out where the line sits plan by plan.
Whizi, which publishes this page, has no API. It's a chat app, and it has a free plan that gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, 2 pictures a month and 5 photo uploads a day, with no card. Pro carries Claude, GPT and Gemini in one chat, and its first week is $1.99. If you need Opus inside your own software, you should keep your Anthropic key. Cheapest way to use Claude Opus compares every route to the model.
- Use Opus 5.5 at $4 in and $20 out unless a test shows an older Opus does your job better
- Count your real prompts with the token counting endpoint before pricing a move from Opus 4.6 or earlier
- Cache any instructions that repeat across requests, since a hit costs 5% of the input price on Opus 5.5
- Send work that can wait to the Batch API for half price
- Leave fast mode off unless speed pays for a doubled price, and never plan on pairing it with Batch
- Move any Opus 4.5 traffic before 24 November 2026, the earliest date Anthropic lists for its retirement
- Set a monthly spend limit in the Claude Console before the first request
Frequently asked questions
How much does the Claude Opus API cost?
Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, as of 14 November 2026. Opus 5, 4.8, 4.7, 4.6 and 4.5 cost $5 and $25. One short Opus 5.5 message costs $0.014 at 1,000 tokens in and 500 out.
How much does Claude Opus cost per token?
Opus 5.5 costs $0.000004 per input token and $0.00002 per output token. Anthropic bills by the million, so that is $4 and $20 per million. The older Opus versions cost $0.000005 and $0.000025 per token.
Is the Claude Opus API free?
No. Anthropic says new users "receive free credits to test the API," then bills actual monthly usage. Anthropic bills the Claude API apart from Claude.ai plans ("the Claude API is billed separately"), so a free chat account is not API credit. Is Claude Opus free covers the chat side.
Is Claude Opus 4.1 still available on the API?
No. Anthropic retired Opus 4.1 on the Claude API on 5 August 2026 and Opus 4 on 15 June 2026. Both can still run on some partner clouds. Anthropic recommends Opus 4.8 as the replacement, and the current price is $4 and $20 on Opus 5.5.
Does the Batch API make Claude Opus cheaper?
Yes. The Batch API takes 50% off both input and output, so Opus 5.5 costs $2 in and $10 out per million tokens, and Opus 5 to 4.5 cost $2.50 and $12.50. Results are not instant, and fast mode cannot be combined with Batch.
When does Claude Opus 4.5 retire?
Anthropic lists the Claude API retirement of Opus 4.5 as "not sooner than November 24, 2026." Anthropic promises at least 60 days' notice before it retires a public model, and names Opus 4.8 as the replacement for the retired Opus 4.1 and Opus 4.