How Much Does Claude API Cost in 2026? Full Pricing Breakdown
The Claude API is pay-as-you-go: as of late 2026, current model rates run from $0.10 input / $0.50 output per million tokens on Haiku 5.5, to $2/$10 on Sonnet 5.5, $4/$20 on Opus 5.5, and $10/$50 on Fable 5.1 — with batch processing cutting token costs 50% and prompt caching discounting repeated input by up to 90%. If you'd rather pay a flat rate, Claude subscriptions run $20/month (Pro, $17/month billed annually) up to $200/month (Max 20x), and Max plans include $100–$200/month in API credits. For small businesses, the practical answer is usually Sonnet 5.5 at $2/$10 per million tokens: a workload of 1M input and 250K output tokens costs about $4.50/month at that rate.
TL;DR
- Claude API pricing is per million tokens: Haiku 5.5 at $0.10/$0.50, Sonnet 5.5 at $2/$10, Opus 5.5 at $4/$20, and Fable 5.1 at $10/$50 (input/output) as of October 2026.
- Batch processing takes 50% off all token rates, and prompt caching cuts repeated input by up to 90% — stacking both can reduce effective spend by 95%+.
- Claude 4.6-and-later models include the full 1M-token context window at standard pricing, with no long-context surcharge (Sonnet 4.5 was the last with one).
- Subscriptions and the API are billed separately: Pro is $20/month ($17/month billed annually), Max is $100–$200/month, and Max plans include $100–$200/month in API credits.
- Team seats cost $20–$25/month (Standard) or $100–$125/month (Premium) with up to $500/month in pooled API credits; Enterprise self-serve is $20/seat/month billed annually (20-seat minimum) plus usage at API rates.
- Extra tools are billed separately: web search at $10 per 1,000 searches, code execution at $0.05 per container-hour (50 free hours per org per day), and Managed Agents runtime at $0.08 per session-hour.
- A sample small-business workload — 1M input and 250K output tokens on Sonnet 5.5 — costs about $4.50/month at standard rates.
Head to head
| Dimension | Claude API |
|---|---|
| Pricing model | Pay-as-you-go per million tokens (input and output billed separately) 🏆 |
| Entry cost | $0.10 input / $0.50 output per 1M tokens (Haiku 5.5) 🏆 |
| Mid-tier rate | Sonnet 5.5 at $2/$10 per 1M tokens; 1M input + 250K output ≈ $4.50 🏆 |
| Discounts | Batch API −50% on all tokens; prompt caching up to −90% on repeated input; stackable to 95%+ 🏆 |
| Context window | 1M tokens at standard pricing on 4.6+ models; Haiku 4.5 capped at 200K 🏆 |
| Setup | Create a Console account, generate an API key, add payment method — no minimum commitment 🏆 |
| Best-fit workload | Variable or spiky usage: product features, extraction, routing, agents with caching 🏆 |
The contenders
The Claude API is Anthropic's developer surface for building on its Claude family of large language models, accessed through the Claude Platform (platform.claude.com, formerly the Claude Console). Unlike the consumer chat app — billed as flat subscriptions (Free, Pro $20/month, Max $100–$200/month, Team and Enterprise seats) — the API is strictly pay-as-you-go: you add a payment method, generate an API key, and are billed per token for input and output. As of October 2026 the current model lineup is Fable 5.1 ($10/$50 per million tokens), Opus 5.5 ($4/$20), Sonnet 5.5 ($2/$10), and Haiku 5.5 ($0.10/$0.50), with 4.x models like Haiku 4.5 ($1/$5) still available. Models from the 4.6 generation onward include a full 1M-token context window at standard pricing. Cost levers include prompt caching (cache reads at a fraction of base input, writes at 1.25x–2x), a Batch API at 50% off, Fast Mode on Opus tiers, and paid server-side tools like web search and code execution. For small businesses, the API suits product features, automation, and agent workloads where usage varies month to month.
pricing 0.1/one_time
- + Entry-level Haiku 5.5 pricing of $0.10/$0.50 per million tokens makes high-volume, simple tasks extremely cheap
- + Batch processing (50% off) and prompt caching (up to 90% off repeated input) stack for savings of 95%+ on eligible workloads
- + 1M-token context included at standard rates on 4.6+ models — no long-context surcharge
- + Transparent published rates per model, with pinned model IDs so snapshots don't silently change
- + Max and Team subscription plans include monthly API credits ($100–$200 for Max, up to $500 pooled for Team), softening the subscription/API split
- − Costs compound on multi-turn agent workloads: a 10-turn chat with a 50K-token history can generate 500K input tokens, mostly repeated context
- − Subscriptions and API usage are billed separately — a Pro or Max plan does not create a shared pool of standard API tokens
- − Fast Mode does not work with the Batch API, limiting discount stacking on latency-sensitive jobs
- − Opus 4.7's new tokenizer could consume up to 35% more tokens for the same text, raising effective costs on migration
- − US-only inference on Claude 4.6 and later adds 10% to rates
Verdict
Winner: Claude API
The Claude API's pay-as-you-go model scales cleanly with usage, and its mid-tier Sonnet 5.5 rate of $2/$10 per million tokens is affordable for most small-business workloads — but costs compound fast on multi-turn, agent-style workloads unless you actively use caching and batch discounts.
| Option | Overall | Breakdown |
|---|---|---|
| Claude API | 8.5 | : 9: 8: 9: 7: 8: 9 |
Claude API token rates in 2026: what each model actually costs
The discounts that change the math: caching, batch, and credits
API vs. subscription: which billing surface fits a small business
Which should you pick?
- :
- :
- :
- :
FAQ
How much does the Claude API cost per token in 2026?
As of October 2026, current-generation Claude API rates per million tokens are: Haiku 5.5 at $0.10 input / $0.50 output, Sonnet 5.5 at $2/$10, Opus 5.5 at $4/$20, and Fable 5.1 at $10/$50. Legacy models remain available, from Claude 3 Haiku at $0.25/$1.25 up to Opus 4.1 at $15/$75. Input and output are billed separately, and rates are per million tokens (roughly 750,000 words on earlier tokenizers).
Is there a minimum spend or subscription fee for the Claude API?
No. API access is pay-as-you-go: you create an account in the Claude Console, generate an API key, add a payment method, and are billed per token. There is no subscription fee or seat minimum for standard API access — Enterprise self-serve ($20/seat/month, 20-seat minimum, plus API-rate usage) is a separate organizational product.
How can I reduce my Claude API bill?
Three main levers. Prompt caching charges 1.25x base input for a 5-minute cache write (2x for 1-hour) but reads back at a fraction of base input — up to 90% off repeated context like system prompts and documents. The Batch API takes 50% off input and output for asynchronous jobs and stacks with caching, for combined savings of 95%+ on eligible workloads. Routing routine tasks to cheaper models (Haiku 5.5 instead of Opus 5.5) is the third lever — a 40x price difference between those tiers.
Does a Claude Pro or Max subscription include API usage?
Subscriptions and API usage are billed separately — a paid plan does not create a shared pool of standard API tokens. That said, Max plans include monthly API credits ($100/month on Max 5x, $200/month on Max 20x), and Team plans include up to $500/month in pooled API credits, so heavier plans do give you a token budget for building. Also note the developer gotcha: if ANTHROPIC_API_KEY is set in your shell, Claude Code bills at API rates and ignores your subscription.
What does a typical small-business Claude API workload cost per month?
It depends heavily on model choice and caching. A concrete example: one million input tokens plus 250,000 output tokens costs about $4.50 on Sonnet 5.5 ($2/$10 rates) and $22.50 on Fable 5.1 ($10/$50). Multi-turn agent workloads cost more because context is re-sent each turn — a 10-turn chat with a 50K-token history can generate 500K input tokens — which is exactly what prompt caching is designed to offset. Add-on tools are billed separately: web search at $10 per 1,000 searches and code execution at $0.05 per container-hour after 50 free daily hours.
Do agent skills work with the Claude API, and how do I pick safe ones?
Agent skills — SKILL.md instruction packages for Claude Code and other agents — are widely shared, but published research has found that roughly 26% of skills carry at least one vulnerability, so vetting matters before you plug them into an API-keyed agent. SkillzBench, a free directory from AIQ Labs, is building exactly this: it crawls and dedupes skills from GitHub, pins each version to a commit SHA, runs security scans and license checks, and benchmarks every skill in a sandbox to publish a measurable Lift. Everything on SkillzBench — the directory, the MCP connector, benchmarks, and alerts — is free, and its trust scores are never influenced by money.