Claude API Pricing (October 2026): Claude Fable 5 and 5.1, Opus, Sonnet, Haiku, Caching, Batch and a Cost Calculator
On this page
Short answer: as of October 17, 2026, Claude Fable 5 and Fable 5.1, Anthropic's top tier, cost $10 / $50 per million input / output tokens. The rest of the Claude API costs $4 / $20 for Opus 5.5, $2 / $10 for Sonnet 5.5 and $0.10 / $0.50 for Haiku 5.5 (prompts up to 100K tokens). Prompt caching cuts repeated input to 2.5–10% of the normal price, and the Batch API halves everything.
Prices below are copied from Anthropic's pricing page on October 15, 2026, and rechecked on October 17, and are in USD per million tokens (MTok). Anthropic changes prices with new model releases, so check the date at the top of this page.
Current models
| Model | Input | Output | 5-min cache write | 1-hour cache write | Cache read |
|---|---|---|---|---|---|
| Claude Fable 5.1 | $10 | $50 | $12.50 | $20 | $0.25 |
| Claude Opus 5.5 | $4 | $20 | $5 | $8 | $0.20 |
| Claude Sonnet 5.5 | $2 | $10 | $2.50 | $4 | $0.10 |
| Claude Haiku 5.5 (prompt ≤ 100K tokens) | $0.10 | $0.50 | $0.125 | $0.20 | $0.01 |
| Claude Haiku 5.5 (prompt > 100K tokens) | $0.50 | $2.50 | $0.625 | $1 | $0.05 |
Which one to pick:
- Sonnet 5.5 is the default for most production work and coding agents.
- Opus 5.5 is for the hardest reasoning and long agentic tasks. Note that it's now only twice the price of Sonnet.
- Haiku 5.5 is for high-volume, simple work: classification, extraction, routing, short replies.
- Fable 5.1 is the top tier, priced at 2.5x Opus 5.5. Details below.
Claude Fable 5 and Fable 5.1 pricing
Fable is Anthropic's most capable generally available model line, for demanding reasoning and long-horizon agentic work. There are two versions on the API:
| Claude Fable 5.1 | Claude Fable 5 | |
|---|---|---|
| Model ID | claude-fable-5-1 |
claude-fable-5 |
| Status | Current, released Sept 1, 2026 | Legacy, released June 9, 2026 |
| Input / output | $10 / $50 | $10 / $50 |
| 5-min / 1-hour cache write | $12.50 / $20 | $12.50 / $20 |
| Cache read | $0.25 | $1 |
| Batch API input / output | $5 / $25 | $5 / $25 |
| Context / max output | 1M / 128K | 1M / 128K |
| Knowledge cutoff | June 2026 | January 2026 |
| Retirement | Not before Sept 1, 2027 | Not before June 9, 2027 |
What this means for your bill:
- Fable 5.1 costs the same as Fable 5 but caches four times cheaper. On a long agent session where most input is cached, 5.1 is the cheaper of the two. Anthropic recommends migrating, but note that moving from Fable 5 to 5.1 has breaking changes, including forced tool use (
tool_choiceofanyor a named tool) returning an error. - Fable always thinks. Thinking is adaptive and can't be turned off, and the default effort is
high, againstmediumon Opus 5.5. Thinking is billed as output at $50 per million tokens, so set a lower effort for simpler tasks. - It's 2.5x the price of Opus 5.5, and Anthropic itself suggests starting with Opus 5.5 for most workloads and moving up to Fable when a task needs it. Fable is also listed as slower.
- Rate limits are shared. Fable 5.1 and Fable 5 draw on one combined limit: 1,000 requests, 500K input tokens and 100K output tokens per minute on the Start tier.
Example: one coding-agent turn with 50K input tokens, 40K of them cache reads, and 5K output tokens:
| Model | Uncached input | Cache reads | Output | Total |
|---|---|---|---|---|
| Fable 5.1 | $0.10 | $0.01 | $0.25 | $0.36 |
| Fable 5 | $0.10 | $0.04 | $0.25 | $0.39 |
| Opus 5.5 | $0.04 | $0.008 | $0.10 | $0.148 |
Fable on Claude plans. Max plans include Fable models at up to 50% of your weekly usage limit. Pro users can use Fable through usage credits, billed separately. Team premium seats include it up to 50% of each member's weekly limit, and the Free plan doesn't include it. For how plans compare with API billing, see Claude Code pricing.
Claude Mythos 5 and 5.1 share Fable's specifications and prices but are only for organizations verified through Anthropic's programs, such as the Cyber Verification Program.
Previous models still on the API
| Model | Input | Output | Cache read |
|---|---|---|---|
| Claude Fable 5 | $10 | $50 | $1 |
| Claude Opus 5 | $5 | $25 | $0.50 |
| Claude Opus 4.8 / 4.7 / 4.6 / 4.5 | $5 | $25 | $0.50 |
| Claude Sonnet 5 | $2 | $10 | $0.20 |
| Claude Sonnet 4.6 / 4.5 | $3 | $15 | $0.30 |
| Claude Haiku 4.5 | $1 | $5 | $0.10 |
Claude Opus 4.1, Opus 4, Sonnet 4 and Haiku 3.5 are retired on the Claude API but still available on some cloud platforms at their old prices.
Sonnet 5 launched with $2 / $10 as introductory pricing through August 31, 2026, with a planned increase to $3 / $15. Anthropic cancelled the increase, so $2 / $10 is now its standard price.
How you're billed
Each request is billed for:
- Input tokens: everything you send, including the system prompt, the conversation history, tool definitions and files.
- Output tokens: everything Claude writes, including its thinking when extended thinking is on.
- Server tools, if you use them. Web search costs $10 per 1,000 searches plus the tokens it adds. Web fetch has no extra charge beyond tokens.
Billing comes out of prepaid credits in USD, and new accounts get a small free credit to test with. See How to get a Claude API key for the setup.
Watch the tokenizer change
Claude 4.7 and later models use a new tokenizer that Anthropic says produces about 30% more tokens for the same text. So a lower per-token price doesn't fully translate into a lower bill. Moving from Sonnet 4.6 ($3 / $15) to Sonnet 5.5 ($2 / $10), the same text costs roughly 2 × 1.3 = $2.60 per "old" million input tokens instead of $3, about 13% cheaper rather than 33%. The exact ratio depends on your content, so measure with Anthropic's token-counting endpoint before you migrate a big workload.
Prompt caching: the biggest lever
If your requests repeat the same prefix (a long system prompt, a document, the conversation so far), cache it. Cached reads are billed at a fraction of the input price:
| Operation | Price vs normal input | Lasts |
|---|---|---|
| 5-minute cache write | 1.25x | 5 minutes, refreshed on each hit |
| 1-hour cache write | 2x | 1 hour |
| Cache read: most models | 0.1x | |
| Cache read: Opus 5.5 and Sonnet 5.5 | 0.05x | |
| Cache read: Fable 5.1 | 0.025x |
A 5-minute cache pays for itself after one hit, and a 1-hour cache after two. Chat apps and coding agents resend the whole conversation every turn, so caching is what keeps long sessions affordable. Claude Code enables it automatically.
Batch API: half price if you can wait
The Message Batches API processes requests asynchronously instead of in real time, at 50% off both input and output. That means Sonnet 5.5 at $1 / $5 and Haiku 5.5 at $0.05 / $0.25. Batch and caching discounts stack. Fast mode isn't available for batches.
Other things that change the price
| Factor | Effect |
|---|---|
| Long context | Claude 4.6+ models (except Haiku 5.5) bill the full 1M-token window at standard rates. A 900K-token request costs the same per token as a 9K one. Haiku 5.5 switches to higher prices above 100K prompt tokens |
| Fast mode | Faster output at premium prices: Opus 5.5 $8 / $40; Opus 5 and Opus 4.8 $10 / $50 (those two in research preview). Claude API, plus Microsoft Foundry for Opus 5.5 |
US-only inference (inference_geo: "us") |
1.1x on all token prices for Claude 4.6+ models |
| Cloud platforms | Amazon Bedrock and Google Cloud set their own prices, and regional endpoints add a 10% premium over global ones |
Worked examples
All three use October 2026 list prices.
1. Customer-support chatbot on Sonnet 5.5. 30,000 requests a month, each with 3,000 input tokens (2,000 of them a cached system prompt) and 400 output tokens:
| Tokens | Price | Cost | |
|---|---|---|---|
| Uncached input | 30M | $2 / MTok | $60 |
| Cached input (reads) | 60M | $0.10 / MTok | $6 |
| Output | 12M | $10 / MTok | $120 |
| Total | ≈ $186 / month |
Without caching, input would cost $180 instead of $66, for about $300 in total.
2. One long coding-agent session on Opus 5.5. 50 turns, an average of 60K tokens of context per turn (3M input tokens in total), about 100K of them newly written to cache across the session, and 1,500 output tokens per turn:
| Tokens | Price | Cost | |
|---|---|---|---|
| Cache reads | ~2.9M | $0.20 / MTok | $0.58 |
| Cache writes (5-min) | ~0.1M | $5 / MTok | $0.50 |
| Output | 75K | $20 / MTok | $1.50 |
| Total | ≈ $2.60 per session |
The same session without caching: 3M × $4 + $1.50 = $13.50, about five times more.
3. Summarizing 100,000 documents with Haiku 5.5 via Batch. 5,000 input and 300 output tokens each: 500M input × $0.05 + 30M output × $0.25 = $32.50.
Cost calculator
Estimate a monthly bill at October 2026 list prices. The calculator ignores the small cache-write premium and server tools.
Rate limits and spend caps
Pricing is only half the picture. New organizations start on lower tiers, each with a monthly spend cap: $500 on Start, $1,000 on Build and $200,000 on Scale. Requests fail with HTTP 429 until the next month once you hit it. You can raise the cap with Request tier increase in the Console. The details are in How to get a Claude API key.
Five ways to spend less
- Cache every long, repeated prefix, such as system prompts, tool definitions and reference documents.
- Route simple work to Haiku 5.5. At $0.10 / $0.50 it's 20x cheaper than Sonnet 5.5.
- Batch anything that can wait an hour or more, for 50% off.
- Cap
max_tokensand ask for concise output. Output tokens cost 5x input on every current model. - Measure per task, not per token. Count tokens on your real prompts, especially when switching tokenizer generations.
Related
FAQ
How much does Claude Fable 5 cost?
Claude Fable 5 and Claude Fable 5.1 both cost $10 per million input tokens and $50 per million output tokens on the Claude API, with the Batch API at half that. The difference is caching. Cache reads cost $1 per million tokens on Fable 5 and $0.25 on Fable 5.1, so Fable 5.1 is cheaper for long sessions with repeated context.
Do Claude Pro or Max plans include Claude Fable?
Max plans include Fable models, using up to 50% of your weekly usage limit. Pro users can use Fable only through usage credits, billed separately. On Team plans, premium seats include Fable up to 50% of each member's weekly limit. The Free plan doesn't include Fable.
How much does the Claude API cost?
As of October 2026, Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, Claude Opus 5.5 costs $4 and $20, and Claude Haiku 5.5 costs $0.10 and $0.50 for prompts up to 100,000 tokens. Claude Fable 5.1, the top model, costs $10 and $50.
How much does prompt caching save?
Cache reads cost 10% of the normal input price on most models, 5% on Opus 5.5 and Sonnet 5.5, and 2.5% on Fable 5.1. Writing to the cache costs 1.25x the input price for a 5-minute cache or 2x for a 1-hour cache, so a 5-minute cache pays for itself after a single cache hit.
Is the Claude Batch API cheaper?
Yes. The Batch API charges 50% of the standard input and output prices for requests you can wait on, and the discount stacks with prompt caching.
Does a long prompt cost more on Claude?
Not on most current models. Claude 4.6 and later (except Haiku 5.5) bill the full 1M-token context window at standard rates. Claude Haiku 5.5 charges higher rates when a prompt exceeds 100,000 tokens.
Is there a free tier for the Claude API?
New accounts receive a small amount of free credit for testing. After that, usage is billed from prepaid credits in USD.
Why does the same text use more tokens on newer Claude models?
Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text, according to Anthropic. Compare models on cost per task, not just price per token.
Found a mistake? Email [email protected].