Requesty
Back|NOV '26AI MODELS / COST OPTIMIZATION
8 MIN READ|

Claude API pricing: model, cache and Batch rates

Last updated

As of November 2026, Claude API pricing starts at $0.10 per million input tokens and $0.50 per million output tokens for Haiku 5.5 with prompts up to 100,000 tokens. Sonnet 5.5 costs $2 and $10, Opus 5.5 costs $4 and $20, and Fable 5.2 costs $8 and $40. Anthropic's Batch API halves input and output rates.

List prices are where a Claude budget starts. The bill also depends on how much text the model reads and writes, how often prompts hit the cache, and whether the work can wait for Batch. Below, we put the rate card next to measured traffic and work through a monthly example.

Claude API prices per million tokens

These are November 2026 standard rates in US dollars. Input tokens are the text you send to the model; output tokens are the text it generates.

ModelInputOutput
Haiku 5.5, up to 100,000 prompt tokens$0.10$0.50
Haiku 5.5, over 100,000 prompt tokens$0.50$2.50
Sonnet 5.5$2$10
Opus 5.5$4$20
Fable 5.2$8$40
Mythos 5.2$8$40
Haiku 4.5$1$5
Sonnet 5$2$10
Opus 5$5$25
Fable 5.1$10$50

Source: Anthropic's pricing documentation.

Haiku 5.5's rate changes once a prompt passes 100,000 tokens: input and output both cost five times as much. Anthropic counts cache reads and writes toward the prompt length, so a cached request can cross the threshold too.

Sonnet 5.5 keeps Sonnet 5's $2 input and $10 output rates, while the cache read rate drops by half, from $0.20 to $0.10 per million tokens. Opus 5.5 costs 20% less per input and output token than Opus 5. Fable 5.2 and Mythos 5.2 cost $8 input and $40 output per million tokens.

How Claude cache pricing works

Prompt caching lets a request reuse a stored prompt prefix. The first request pays for a cache write, and later requests can pay the lower cache read rate for the reused text.

Anthropic's cache rules set a five-minute write at 1.25 times the standard input rate and a one-hour write at twice the input rate. The November 2026 read rates below follow Anthropic's published model pricing and cache read rules; the write rates apply those multipliers.

ModelCache read per million tokensFive-minute write per million tokensOne-hour write per million tokens
Haiku 5.5, up to 100,000 prompt tokens$0.01$0.125$0.20
Haiku 5.5, over 100,000 prompt tokens$0.05$0.625$1
Sonnet 5.5$0.10$2.50$4
Opus 5.5$0.20$5$8
Fable 5.2$0.20$10$16
Mythos 5.2$0.20$10$16
Haiku 4.5$0.10$1.25$2
Sonnet 5$0.20$2.50$4
Opus 5$0.50$6.25$10
Fable 5.1$0.25$12.50$20

With Opus 5.5, writing one million tokens to a five-minute cache costs $5. Reading those tokens once adds $0.20, for $5.20 across the two requests. Sending the same million tokens as standard input twice would cost $8, so the write pays off after one full reuse within the cache period.

Caching also has a minimum prompt length. Anthropic sets it at 512 tokens for Fable 5.2, Fable 5.1, Opus 5.5, Opus 5, Sonnet 5.5 and Haiku 5.5; Sonnet 5 has a 1,024-token minimum. For rates beyond Claude, see our cache read price comparison across providers.

What else changes the token bill

Batch is useful for work that can wait. At Opus 5.5's November 2026 standard rates, a job with one million input tokens and 500,000 output tokens costs $14: $4 for input plus $10 for output. The same job costs $7 on Batch.

Pricing modifierEffect
Batch APIHalves input and output token rates. Cache multipliers can also apply.
Opus 5.5 fast modeCosts $8 per million input tokens and $40 per million output tokens. Fast mode cannot be combined with Batch.
Fable 5.2 and Mythos 5.2 fast modeCosts $16 per million input tokens and $80 per million output tokens.
US-only inferenceAdds a 1.1x multiplier to all token pricing categories for Claude 4.6 and later models.

Sources: Anthropic's pricing documentation and fast mode documentation.

There's another cost to watch when comparing models: Anthropic says the tokenizer used by Claude 4.7 and later produces approximately 30% more tokens for the same text. A token is a chunk of text, and different tokenizers can split the same text differently. Compare the cost of a sample of your own requests alongside the price per million tokens.

What Claude costs in Requesty traffic

Requesty's production rankings cover traffic from September 14 to November 16, 2026. The blended rate is the price billed per million tokens across requests, including cache discounts and the observed input and output mix.

ModelStandard input per millionBlended cost per millionCache hit rateBill removed by cachingShare of tokensShare of spend
Opus 5.5$4$1.1885%80.4%9.4%26.4%
Sonnet 5.5$2$1.0781.1%69.5%2.9%7.3%
Haiku 5.5$0.10 for prompts up to 100,000 tokens$0.1273.2%72.9%2.1%0.6%
Fable 5.1$10$2.1687.7%88.2%0.4%1.9%

A blended rate includes both input and output, so it cannot be compared directly with the standard input column. It gives a view of the bill for the traffic in the rankings. Your rate will depend on your own token mix and cache usage.

Take Opus 5.5: it accounted for 9.4% of tokens and 26.4% of spend. Sonnet 5.5's standard input and output rates are half of Opus 5.5's, yet its measured blended rate was $1.07 per million tokens against $1.18 for Opus. The traffic sent to each model and its cache behavior affect that comparison. Across all 40 models in the rankings, caching removed 74.9% of the bill and the blended rate was $0.42 per million tokens.

Worked example: 100 million tokens a month

Take a workload of 100 million tokens per month, using the same blended rates and cache savings shares as the Requesty traffic above. Those shares differ by model, so each estimate is based on that model's observed traffic.

To estimate the bill without caching, divide the blended rate by the share of the bill left after caching. For Opus 5.5, that is $1.18 divided by 0.196, or about $6.02 per million tokens. Multiplying each rate by 100 gives the monthly amounts.

An 80/20 input and output split gives another way to check the scale of the estimate. At November 2026 list prices, one million Opus 5.5 tokens with that split costs $7.20 ($4 times 0.8, plus $20 times 0.2). This list-price check excludes caching and uses an assumed token mix.

ModelBlended cost per millionMonthly cost with cachingEstimated monthly cost without cachingEstimated monthly saving
Opus 5.5$1.18$118$602$484
Sonnet 5.5$1.07$107$351$244
Fable 5.1$2.16$216$1,831$1,615
Haiku 5.5$0.12$12$44$32

Source: Requesty rankings. Uncached rates and monthly totals are derived from the published blended rates and cache savings shares.

Cost per solved coding task

A lower token bill can still buy fewer completed tasks. Requesty's coding benchmark tested models on 129 private Go backend tasks, with results updated October 29, 2026.

ModelTasks solvedCost per taskCost per solved task
Opus 5.592.2%$0.93$1.01
Fable 5.189.1%$3.48$3.90
Sonnet 5.587.6%$0.52$0.59
Haiku 5.581.4%$0.11$0.14

Opus 5.5 solved the largest share of tasks in this group. Its standard input price is 40 times Haiku 5.5's rate for prompts up to 100,000 tokens, while its benchmark cost per solve is about seven times Haiku's. The benchmark measures one set of Go backend tasks, so run the same comparison on your own workload before choosing a model. Our cost per task analysis looks at how this comparison changes across a wider benchmark suite.

Verdict

Start with Haiku for simpler tasks and Sonnet for everyday production work. Test Opus on harder reasoning and coding tasks where a higher solve rate could justify the extra cost. Anthropic gives the same broad tier guidance in its pricing documentation.

Which Claude plans include API usage?

Pro is an app subscription and does not include API usage. Max includes a $100 monthly API credit, Ultra includes a $200 monthly API credit, and Enterprise usage is billed at API rates, according to Anthropic's pricing page. Anthropic also has customized Enterprise and API pricing for volume and committed spend.

PlanPrice
Free$0
Pro$20 per month, or $200 a year billed up front
Max$100 per month, including a $100 monthly API credit
Ultra$200 per month, including a $200 monthly API credit
Enterprise$20 per seat per month, plus usage billed at API rates

Check your Claude bill

Start with a sample of requests from the task you want to run. Record input and output tokens, cache hits and cost per completed task; then compare Haiku, Sonnet and Opus on the same sample.

Requesty has one OpenAI-compatible API across providers, with cost analytics by model and team, budget caps and per-user spend limits. You can bring your own provider keys or use one Requesty key, and route simpler tasks to a cheaper model. Requesty charges 5% on top of model costs, with all features included. See the Requesty rankings for measured model costs, then use the AI gateway to compare routes for your workload.

Frequently asked questions
How much does the Claude API cost?
In November 2026, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Sonnet 5.5 costs $2 input and $10 output, Opus 5.5 costs $4 and $20, and Fable 5.2 costs $8 and $40 per million tokens. Batch halves those rates.
What is Claude Opus 5.5 API pricing per million tokens?
In November 2026, Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens at standard rates. Cache reads cost $0.20 per million tokens, while Batch halves the standard input and output rates.
How much does Claude Sonnet 5.5 cost?
In November 2026, Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens. Cache reads cost $0.10 per million tokens.
How much does Claude prompt caching save?
On Requesty traffic from September 14 to November 16, 2026, caching removed 80.4% of the Claude Opus 5.5 bill and 69.5% of the Claude Sonnet 5.5 bill.
Does a Claude Pro subscription include API usage?
Claude Pro covers app usage and does not include API usage. Max includes a $100 monthly API credit and Ultra includes a $200 monthly API credit, while Enterprise usage is billed at API rates.
Does the Claude Batch API reduce token prices?
In November 2026, Anthropic's Batch API halves input and output token prices. A Claude Opus 5.5 job with 1 million input tokens and 500,000 output tokens costs $7 on Batch instead of $14 at standard rates.
Related reading

Start building with Requesty

One line of code. 600+ models. Full control.

Speak to founders