Key takeaways
- Claude API pricing is pay-as-you-go per million tokens (MTok), split into input and output. In 2026 it ranges from $0.10 / $0.50 (Haiku 5.5) to $8 / $40 (Fable 5.2).
- Opus 5.5 is the default pick at $4 / $20. Anthropic recommends starting there for most workloads.
- Thinking tokens are billed as output. You pay for the full internal thinking, even when the response shows only a summary. The
effortsetting is your main cost lever. - Prompt caching is the biggest saving. Cache reads cost 2.5%–10% of the input price, so repeated context becomes very cheap.
- The full 1M context window is billed at standard rates on every current model except Haiku 5.5, which charges more for prompts over 100,000 tokens.
- A Claude subscription and the API are billed separately. Pro, Max and Team cover the Claude apps and Claude Code; the API charges for each token you use.
Claude API pricing at a glance
Official rates for the current Claude models, in USD per million tokens:
| Model | Input | Output | Cache read | Best for |
|---|---|---|---|---|
| Claude Fable 5.2 | $8 | $40 | $0.20 | Hardest reasoning, long-horizon agents |
| Claude Opus 5.5 | $4 | $20 | $0.20 | Default for coding and knowledge work |
| Claude Sonnet 5.5 | $2 | $10 | $0.10 | Well-scoped coding and multi-step work |
| Claude Haiku 5.5 | $0.10 | $0.50 | $0.01 | High-volume, low-latency tasks |
Haiku 5.5 prices apply to prompts up to 100,000 tokens. All four models support a 1M-token context window and 128K output tokens per request.
All Claude models and their API prices
The full price list, including earlier models that are still available:
| Model | Input | 5-min cache write | 1-hour cache write | Cache read | Output |
|---|---|---|---|---|---|
| Claude Fable 5.2 | $8 | $10 | $16 | $0.20 | $40 |
| Claude Fable 5.1 | $10 | $12.50 | $20 | $0.25 | $50 |
| Claude Fable 5 | $10 | $12.50 | $20 | $1 | $50 |
| Claude Opus 5.5 | $4 | $5 | $8 | $0.20 | $20 |
| Claude Opus 5 | $5 | $6.25 | $10 | $0.50 | $25 |
| Claude Opus 4.8 / 4.7 / 4.6 / 4.5 | $5 | $6.25 | $10 | $0.50 | $25 |
| Claude Sonnet 5.5 | $2 | $2.50 | $4 | $0.10 | $10 |
| Claude Sonnet 5 | $2 | $2.50 | $4 | $0.20 | $10 |
| Claude Sonnet 4.6 / 4.5 | $3 | $3.75 | $6 | $0.30 | $15 |
| Claude Haiku 5.5 (prompts ≤100K) | $0.10 | $0.125 | $0.20 | $0.01 | $0.50 |
| Claude Haiku 5.5 (prompts >100K) | $0.50 | $0.625 | $1 | $0.05 | $2.50 |
| Claude Haiku 4.5 | $1 | $1.25 | $2 | $0.10 | $5 |
Notes:
- Sonnet 5 stays at $2 / $10. Its launch price was announced as introductory, but Anthropic has made it the standard price and dropped the planned increase to $3 / $15.
- Mythos 5.2, 5.1 and 5 share the prices of the Fable model with the same version number and are offered to vetted organizations.
- Sonnet 4.5 retires on November 30, 2026. Plan a move to Sonnet 5.5, which costs less per token.
How Claude API billing works
Every request is billed by the tokens it uses:
- Input tokens: your system prompt, messages, tool definitions, images and documents.
- Output tokens: the response plus all thinking. Anthropic bills "the full thinking process, not the thinking content visible in the response."
- Cache writes and reads: billed at their own rates when you use prompt caching.
Three details change the bill more than people expect:
- Thinking counts as output. Current models think adaptively, and depth is set by
effort(low,medium,high,xhigh,max). Fable 5.2 and Sonnet 5.5 default tohigh; Opus 5.5 and Haiku 5.5 default tomedium. A lower effort level is the fastest way to cut output spend on routine tasks. - Newer models count more tokens. Claude 4.7 and later use a tokenizer that produces roughly 30% more tokens for the same text. On these models, 1M tokens is about 555,000 English words.
- Earlier thinking is billed as input. Thinking blocks from previous turns that stay in context are billed as input tokens on later turns.
The usage object in each response shows the authoritative counts. output_tokens already includes thinking, and output_tokens_details.thinking_tokens shows the thinking share.
How much do 1 million Claude API tokens cost?
| Model | 1M input tokens | 1M output tokens | 1M tokens at a 3:1 input/output mix |
|---|---|---|---|
| Claude Fable 5.2 | $8.00 | $40.00 | $16.00 |
| Claude Opus 5.5 | $4.00 | $20.00 | $8.00 |
| Claude Sonnet 5.5 | $2.00 | $10.00 | $4.00 |
| Claude Haiku 5.5 | $0.10 | $0.50 | $0.20 |
The 3:1 column assumes 750K input and 250K output tokens, a common shape for chat and coding work. Output-heavy tasks such as long-form writing cost closer to the output column.
Prompt caching: the biggest saving
Prompt caching stores a repeated prefix, such as a long system prompt, a codebase or a document, so later requests read it at a fraction of the input price.
| Cache operation | Price | Pays off after |
|---|---|---|
| 5-minute cache write | 1.25× base input | One cache read |
| 1-hour cache write | 2× base input | Two cache reads |
| Cache read: Fable 5.2 and Fable 5.1 | 0.025× base input | — |
| Cache read: Opus 5.5 and Sonnet 5.5 | 0.05× base input | — |
| Cache read: all other models | 0.1× base input | — |
Caching multipliers also apply together with data residency pricing. For agents that re-read the same repository on every turn, caching often removes most of the input bill.
Fast mode
Fast mode delivers up to 2.5× faster output at premium prices:
| Model | Fast mode input | Fast mode output |
|---|---|---|
| Claude Fable 5.2 | $16 | $80 |
| Claude Opus 5.5 | $8 | $40 |
| Claude Opus 5 | $10 | $50 |
Fast mode applies across the full context window and stacks with caching and data residency.
Long context, data residency and cloud platforms
- 1M context at standard prices. Claude 4.6 and later models bill the full 1M-token window at their normal rates.
- Haiku 5.5 over 100K tokens. A prompt over 100,000 tokens is billed at $0.50 / $2.50 for the whole request, including any cached part.
- US-only inference. Setting
inference_geoto"us"on Claude 4.6 and later applies a 1.1× multiplier to every token category. The default global routing uses standard prices. - Cloud platforms. Regional endpoints on Amazon Bedrock and Google Cloud carry a 10% premium over global endpoints for Sonnet 4.5, Haiku 4.5, Opus 4.5 and later models.
What tools cost
| Tool | Price |
|---|---|
| Web search | $10 per 1,000 searches, plus tokens for the results |
| Web fetch | Standard token costs for the fetched content |
| Code execution | Free with web search or web fetch; otherwise 1,550 free hours per organization per month, then $0.05 per container-hour |
| Bash tool | About 325 extra input tokens per request (Claude 4.7 and later) |
| Text editor tool | About 974 extra input tokens per request (Claude 4.7 and later) |
| Computer use toolset | About 4,500 extra input tokens with default members |
Tool definitions and tool-use system prompts are billed as input tokens on every request that includes them, so keep the tool list to what each request needs.
Real Claude API cost examples
All examples use official standard prices.
| Scenario | Model | Usage | Cost |
|---|---|---|---|
| Support chatbot, 100,000 conversations a month | Haiku 5.5 | 2,000 input + 300 output tokens each | 200×0.10 + 30×0.50 = $35 / month |
| Coding agent session with a cached repository | Opus 5.5 | 2M cache reads + 300K fresh input + 150K output | 2×0.20 + 0.3×4 + 0.15×20 = $4.60 |
| Document pipeline, 10,000 files | Sonnet 5.5 | 8,000 input + 1,000 output tokens each | 80×2 + 10×10 = $260 |
| Hard reasoning task | Fable 5.2 | 200K input + 40K output | 0.2×8 + 0.04×40 = $3.20 |
In the chatbot row, 100,000 conversations × 2,000 tokens = 200M input tokens, which is why the formula multiplies by 200.
Which Claude model fits your budget?
| Budget priority | Start with | Why |
|---|---|---|
| Lowest cost at scale | Haiku 5.5 | $0.10 / $0.50, keep prompts under 100K tokens |
| Balanced cost for coding and agents | Sonnet 5.5 | Half the price of Opus 5.5, cheapest cache reads after Haiku |
| Best quality per dollar for most work | Opus 5.5 | Anthropic's recommended default |
| Hardest problems | Fable 5.2 at low or medium effort | Anthropic reports it matching Opus 5.5's best scores for a fraction of the cost per task |
For a deeper look at the top two models, see Claude Fable 5.2 vs Opus 5.5.
Claude API vs Claude subscription (Pro, Max, Team)
Claude subscriptions cover the Claude apps and Claude Code with usage limits. The API is metered per token for your own applications.
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | Limited usage of Sonnet and Haiku in the Claude apps |
| Pro | $20 / month ($17 / month billed annually) | Opus, Claude Code, Projects, usage credits |
| Max | $100 / month | 5× Pro usage, plus $100 a month in API credits |
| Ultra | $200 / month | 10× Pro usage, plus $200 a month in API credits |
| Team | $25 / seat / month ($20 annually); premium seat $125 ($100 annually) | 2–150 seats, up to $500 a month in pooled API credits |
| Enterprise | $20 / seat / month billed annually, plus usage at API rates | 20-seat minimum |
Plan prices are from claude.com/pricing and exclude tax.
Which should you choose? A subscription suits people who work in Claude or Claude Code every day and want a predictable bill. The API suits applications, automation and bursty usage, where you pay only for tokens used and can lower costs with caching and effort settings.
Claude vs OpenAI vs Gemini API pricing
Standard prices per million tokens on November 14, 2026:
| Model | Input | Output |
|---|---|---|
| Claude Fable 5.2 | $8 | $40 |
| Claude Opus 5.5 | $4 | $20 |
| Claude Sonnet 5.5 | $2 | $10 |
| Claude Haiku 5.5 | $0.10 | $0.50 |
| GPT-6 Astra | $10 | $50 |
| GPT-6.1 Sol (input ≤272K) | $2 | $10 |
| GPT-6.1 Terra (input ≤272K) | $1 | $5 |
| GPT-6 Luna (input ≤272K) | $0.10 | $0.50 |
| Gemini 4 Argon (through November 30, 2026) | $2 | $10 |
| Gemini 4 Argon (from December 1, 2026) | $4 | $20 |
All three providers bill reasoning or thinking tokens as output. OpenAI raises GPT-6.x prices for inputs over 272K tokens, while Claude keeps standard rates across its 1M window (except Haiku 5.5 over 100K). Compare cost per completed task on your own workload, since models use different numbers of tokens for the same job. For model-level comparisons, see GPT-6.1 Sol vs Claude Opus 5.5 and our Gemini 4 Argon API guide.
Is the Claude API free?
New Claude API accounts receive free credits to test the API. After that, usage is pay-as-you-go. The free Claude plan covers the Claude apps and is billed separately from the API.
How to lower your Claude API bill
- Match the model to the task. Route simple, high-volume work to Haiku 5.5 or Sonnet 5.5 and keep Opus 5.5 or Fable 5.2 for harder requests.
- Set
effortexplicitly. Since thinking is billed as output,lowormediumcuts costs on routine prompts. - Cache stable prefixes. System prompts, tool definitions and shared documents are ideal cache targets.
- Use PoYo's lower per-token prices. Every current Claude model costs at least 20% less on PoYo than at official rates, through the same API format.
- Cap
max_tokensto bound runaway outputs. - Keep Haiku 5.5 prompts under 100K tokens to stay on its lower price tier.
- Trim tool lists. Each tool definition adds input tokens to every request.
Claude API pricing on PoYo
PoYo gives you every current Claude model through one API key, with pay-as-you-go credits ($1 = 200 credits) and prices below Anthropic's official rates:
| Model | Official input / output | PoYo input / output |
|---|---|---|
| Claude Fable 5.2 | $8 / $40 | $6.40 / $32 |
| Claude Opus 5.5 | $4 / $20 | $3.20 / $16 |
| Claude Sonnet 5.5 | $2 / $10 | $1.60 / $8 |
| Claude Haiku 5.5 | $0.10 / $0.50 | $0.08 / $0.40 |
| Claude Opus 5 | $5 / $25 | $2 / $10 |
| Claude Sonnet 5 | $2 / $10 | $0.85 / $4.275 |
PoYo supports the Anthropic Messages API and an OpenAI-compatible endpoint. With the Anthropic SDKs, set the base URL to https://api.poyo.ai:
curl https://api.poyo.ai/v1/messages \
-H "x-api-key: $POYO_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-opus-5-5",
"max_tokens": 4000,
"output_config": {"effort": "medium"},
"messages": [{"role": "user", "content": "Summarize this contract in five bullet points: ..."}]
}'
Change "model" to claude-fable-5-2, claude-sonnet-5-5 or claude-haiku-5-5 to switch models with the same key.
FAQ
How much is 1 million API tokens on Claude?
It depends on the model and on whether the tokens are input or output. One million input tokens cost $0.10 on Haiku 5.5, $2 on Sonnet 5.5, $4 on Opus 5.5 and $8 on Fable 5.2. One million output tokens cost five times as much: $0.50, $10, $20 and $40.
How much does a Claude API call cost?
Multiply each token count by its rate. A typical Opus 5.5 call with 5,000 input and 1,000 output tokens costs 0.005×4 + 0.001×20 = $0.04. The same call costs $0.001 on Haiku 5.5 and $0.08 on Fable 5.2.
What is Claude Pro?
Claude Pro is Anthropic's $20-a-month subscription ($17 a month billed annually) for the Claude apps. It includes Opus, Claude Code and Projects with higher usage limits than the free plan. It is separate from API billing.
What is the difference between Claude Pro and Max for Claude Code?
Both plans include Claude Code. Max costs $100 a month and gives 5× Pro usage, plus $100 a month in API credits; Ultra costs $200 a month and gives 10× Pro usage, plus $200 a month in API credits. Heavy daily Claude Code users usually outgrow Pro limits first.
Are thinking tokens billed on the Claude API?
Yes, as output tokens. You pay for the full internal thinking, even if the response shows a summary. Lower effort levels reduce thinking and cost.
Does Claude charge extra for long context?
Current Claude models bill the full 1M-token context window at standard prices. Haiku 5.5 is the one exception: prompts over 100,000 tokens use its higher price tier.
Is the Claude API cheaper than OpenAI?
It depends on the tier. Sonnet 5.5 ($2 / $10) matches GPT-6.1 Sol's short-context price, Haiku 5.5 matches GPT-6 Luna ($0.10 / $0.50), and Opus 5.5 ($4 / $20) costs less than GPT-6 Astra ($10 / $50). For inputs over 272K tokens, Claude's flat 1M-context pricing often comes out ahead.