match report

Claude API pricing (2026): what each Claude model costs per token, and how to pay less

updated 2026-10-31 · prices checked on vendor pages

quick answer: Claude API pricing, as of oct 31, 2026, per million tokens (input / output): Sonnet 5.5 is $2 / $10, Opus 5.5 is $4 / $20, Haiku 5.5 is $0.10 / $0.50 for prompts up to 100,000 tokens, and Fable 5.1 is $10 / $50. you pay for what you use, billed monthly in USD. there's no free plan, just a small amount of free credits for new users. the batch API halves every price, and cache hits cost a fraction of normal input.

the Anthropic Claude API is #1 of 10 on our LLM APIs & models board (9.0/10). not because it's cheapest (it isn't) but because it's the one to beat for coding agents and tool use: if the thing you're building edits files, calls tools and has to get it right on the third step, this is where we'd put the key. price is the second question. here's the answer to it anyway.

the price per model

as of oct 31, 2026, from Anthropic's pricing page, per million tokens:

Claude Mythos 5.1 is priced like Fable 5.1, but it's limited availability: only organisations verified through Anthropic's programmes (such as its cyber verification program) get it. older models are still listed: Opus 5 at $5 / $25, Sonnet 5 at $2 / $10 (its launch price is now the standard price), Sonnet 4.6 at $3 / $15 and Haiku 4.5 at $1 / $5. so Opus 5.5 is cheaper per token than Opus 5.

which model should you pay for?

• coding agents and anything with lots of tool calls: Sonnet 5.5. $2 / $10 is the sweet spot, and it's what most of your bill should run on.
• the hard bits (big refactors, planning, gnarly debugging): Opus 5.5 at $4 / $20. twice Sonnet, so route the hard steps there, not every step.
• classifying, extracting, cheap sub-agents: Haiku 5.5 at $0.10 / $0.50. keep prompts under 100,000 tokens or the price jumps five-fold for that request.
• the top of the range: Fable 5.1 at $10 / $50. five times Sonnet. pay it when you've measured that you need it, not because it's the biggest number.

a rough sum, our arithmetic on the list prices: an agent run that reads 1 million input tokens and writes 100,000 output tokens costs about $3 on Sonnet 5.5 ($2 + $1), $6 on Opus 5.5 and $0.15 on Haiku 5.5 (if no single request goes over 100K tokens). caching can knock a big chunk off the input side.

how to pay less

• prompt caching. a cache hit costs 0.1x the input price on most models, 0.05x on Opus 5.5 and Sonnet 5.5, and 0.025x on Fable 5.1 and Mythos 5.1. writing to the 5-minute cache costs 1.25x input, the 1-hour cache 2x. for an agent that resends the same system prompt and repo context every turn, this is the biggest saving going.
• the batch API. 50% off input and output for jobs that can wait. stacks with caching.
• the right model per step. Haiku for the grunt work, Sonnet for the loop, Opus only for the steps that need it.
• mind the tokenizer. Anthropic says Claude 4.7 and later models use a newer tokenizer that makes roughly 30% more tokens for the same text. so a per-token price that looks the same as an older model can still mean a bigger bill.

the extras that show up on the bill

• long context: Claude 4.6 and later models (except Haiku 5.5) get the full 1M token context window at standard prices, so a 900K-token request is billed at the same rate as a 9K one.
• fast mode: faster output on Opus 5.5 for $8 / $40 per million tokens. not available with the batch API.
• US-only inference: setting inference_geo to us adds a 1.1x multiplier to every token price on Claude 4.6 and later. the global default is standard price.
• web search: $10 per 1,000 searches, plus the tokens the results add. web fetch costs only the tokens.
• code execution: free when used with web search or web fetch. on its own, each organisation gets 1,550 free hours a month, then $0.05 an hour per container.
• tool use overhead: turning on tools adds a system prompt of a few hundred tokens per request (286 on Sonnet 5.5 and Opus 5.5), and tool definitions count as input.
• Claude Managed Agents: normal token prices plus $0.08 per session-hour while a session is running. idle time isn't billed.

is the Claude API free?

no. Anthropic says new users get a small amount of free credits to test the API, and doesn't put a number on it. after that it's pay as you go: you buy credits by card in the Claude Console, billing is based on actual monthly usage, and enterprise customers can arrange invoicing. rate limits rise through usage tiers as you spend more. volume discounts are negotiated case by case.

if you want Claude in your terminal for a flat monthly fee rather than per token, that's a Claude subscription with Claude Code, not the API: see Claude Code pricing.

Claude API vs the other LLM APIs

• the value pick with a free tier: Google Gemini API, #2 (8.5/10). see Gemini API pricing
• the biggest ecosystem: OpenAI API, #3 (8.5/10)
• many models on one key, Claude included: OpenRouter, #4. list price plus a fee on credits, see OpenRouter pricing
• much cheaper per token: DeepSeek API, #6 (8.0/10)

the rest of the field

faq

how much does the Claude API cost?

per million tokens, as of oct 31, 2026: Sonnet 5.5 $2 input / $10 output, Opus 5.5 $4 / $20, Haiku 5.5 $0.10 / $0.50 (prompts up to 100K tokens), Fable 5.1 $10 / $50. the batch API is half price.

is the Claude API free?

no. new users get a small amount of free credits to test it, then you pay per token.

how much is Claude Opus 5.5 on the API?

$4 per million input tokens and $20 per million output tokens. cache hits are $0.20, batch is $2 / $10, and fast mode is $8 / $40.

does prompt caching save money on the Claude API?

yes. a cache hit costs 10% of the input price on most models (5% on Opus 5.5 and Sonnet 5.5). a 5-minute cache write costs 1.25x input, so it pays off after one cache read.

is the Claude API the same as a Claude Pro subscription?

no. the API is billed per token through the Claude Console. Claude Pro and Max are monthly subscriptions for the Claude apps and Claude Code. see Claude Code pricing.

tools in this guide

• Anthropic Claude API
• Google Gemini API
• OpenAI API
• OpenRouter
• DeepSeek API

sources

checked 2026-10-31

• Anthropic Claude API pricing (per-model input, output, cache write and cache hit prices, Haiku 5.5 100K-token prompt threshold, Mythos limited availability, older models, Sonnet 5 standard price, caching multipliers, batch API 50% and batch prices, tokenizer note, long context, fast mode, data residency 1.1x, web search $10 per 1,000, web fetch, code execution 1,550 free hours then $0.05/hour, tool use system prompt tokens, Managed Agents $0.08 per session-hour, free credits for new users, monthly usage billing, USD, cards and invoicing, usage tiers, volume discounts): https://platform.claude.com/docs/en/about-claude/pricing
• ranks, scores, verdicts and plans: numptr data/tools.json (LLM APIs & models board)