Claude API Pricing per Million Tokens: Every Model, Cache and Batch Rate
Claude API pricing runs from $0.10 per 1M input tokens on Haiku 5.5 to $8 on Fable 5.2, with output always about 5x the input price, cache reads at a fraction of input, and a 50% discount for batch jobs returned within 24 hours.
Last updated: November 2026
Key takeaways
- Fable 5.2 is $8 input and $40 output per 1M tokens (Anthropic); other rates below are from TrueFoundry's published breakdown, so confirm on Anthropic's page before you budget.
- Output costs 5x input on every model listed.
- Cache writes cost 1.25x the input rate and cache reads cost roughly a tenth or less.
- The Batch API is 50% cheaper for work returned within 24 hours and stacks with caching.
- Prompt caching cut a Sonnet 5.5 support bot from about $4,500 to $2,600 a month in the source's example.
What does each Claude model cost per million tokens?
| Model | Input $/1M | Output $/1M | Cache read $/1M | Cache write $/1M | |---|---|---|---|---| | Fable 5.2 | $8 | $40 | $0.20 | $10 (5 min) | | Fable 5.1 | $10 | $50 | $0.25 | $12.50 | | Opus 5.5 | $4 | $20 | $0.20 | $5.00 | | Sonnet 5.5 | $2 | $10 | $0.10 | $2.50 | | Haiku 5.5 (prompt up to 100K) | $0.10 | $0.50 | $0.01 | $0.125 | | Haiku 5.5 (prompt over 100K) | $0.50 | $2.50 | $0.05 | $0.625 |
Sources: TrueFoundry's Claude API pricing breakdown for all rows except Fable 5.2, which comes from Anthropic's launch post and Latent Space for cache rates. Note the Haiku 5.5 tier jump: a prompt over 100K tokens costs 5x more per token.
What multipliers and discounts apply?
- Cache write: 1.25x the input rate, paid once per write.
- Cache read: about 0.1x of input or less (0.05x on Opus 5.5 and Sonnet 5.5, per TrueFoundry).
- Batch API: 50% off for results within 24 hours, stackable with caching.
- US-only inference: 1.1x standard pricing.
- Opus 5.5 fast mode: 2x. Fable 5.2 fast mode is $16 input and $80 output.
See our deeper reads on Claude Sonnet 5 pricing and margins and Claude Opus 5 cost per task.
How much does Claude cost for a real SaaS workload?
These are the source's worked monthly examples, real-time pricing:
| Workload | Model | No caching | With caching | Saving | |---|---|---|---|---| | Support chat, 500,000 turns (2,000-token cached prefix, 1,000 fresh input, 300 output) | Sonnet 5.5 | about $4,500 | about $2,600 | about 42% | | RAG feature, 100,000 requests (50,000-token context, 1,000 output) | Sonnet 5.5 | about $11,000 | about $2,450 (45,000-token cached prefix, before write overhead) | about 78% | | Agent loop, 240,000 steps (25,000 context, 600 output) | Sonnet 5.5 | about $13,440 | about $3,400 (22,000-token cached prefix) | about 75% |
The same agent loop on Opus 5.5 without caching is about $26,880, exactly 2x Sonnet. Our verification of the RAG row: 50,000 input tokens at $2 per 1M is $0.10 plus $0.01 for 1,000 output tokens, so $0.11 a request and $11,000 for 100,000 requests.
Which Claude model should you pick?
Start with the cheapest model that passes your eval. Haiku 5.5 for extraction, routing and classification; Sonnet 5.5 for most product features; Opus 5.5 for harder reasoning; Fable 5.2 only where nothing else works, since it costs 2x Opus 5.5 per token. Compare vendors on the Anthropic vs OpenAI page and model a support bot in the AI chatbot use case.
Frequently asked questions
How much does the Claude API cost per million tokens?
From $0.10 input and $0.50 output on Haiku 5.5 (prompts up to 100K) to $8 input and $40 output on Fable 5.2. Opus 5.5 is $4 and $20, and Sonnet 5.5 is $2 and $10 per TrueFoundry's table.
How much does prompt caching save on Claude?
In the source's examples, caching cut a support bot's cost by about 42%, a RAG feature's by about 78% and an agent loop's by about 75%, depending on how much of the prompt is a stable prefix.
Is there a batch discount for Claude?
Yes. The Batch API is 50% cheaper for work returned within 24 hours, and it stacks with prompt caching.
Why is output so much more expensive than input?
Output tokens are generated one at a time and cost about 5x the input price on every model listed, so long answers move your bill more than long prompts. The takeaway: your Claude bill is set by model choice, cache hit rate and output length far more than by any single list price. Plug your own mix into the Calcaas LLM cost calculator. Note: place the JSON-LD above inside a <script type="application/ld+json"> tag in the page head.
More from the blog
LLM EconomicsFixed-Price vs Pay-As-You-Go Inference: Breakeven Utilization
Fixed-price inference plans beat pay-as-you-go only once your traffic fills roughly 60 to 80% of the monthly token allowance; the breakeven utilization is u = P / (A x r), the plan price divided by what the same allowance would cost metered.
LLM EconomicsClaude Prompt Caching Break-Even: The Math Behind the Savings
Claude prompt caching breaks even on the second request: a cache write costs 1.25x the base input price, a read costs 0.1x (0.05x on Opus 5.5 and Sonnet 5.5), so a single reuse inside the 5-minute window repays the write premium.
LLM EconomicsFable 5.2 Cost per Task: Cheaper Tokens, a 31% Higher Bill
Fable 5.2 cut list prices 20% to $8 input and $40 output per 1M tokens, yet SemiAnalysis measured its cost per task rising 31%, from $67.16 to $88.15, because the model uses far more tokens to check its own work.
Pricing math, in your inbox.
One short note a week on AI pricing, token economics, and margin. No spam, unsubscribe anytime.