Claude API Cost and Anthropic Plan Pricing
Claude API billing and Claude subscriptions solve different needs. API usage is metered by input and output tokens, while Pro, Max, Team, and Enterprise plans cover the Claude application. The live table below focuses on API channels; linked plan pages cover subscription entitlements.
Current API price snapshot
Up to 12 tracked language models are shown. Channels are selected by lowest input rate normalized to USD, not lowest total cost. Examples use 1M input plus 250K output tokens, without additional cache, batch or context-tier adjustments or tool fees. Processing-mode variants are not interchangeable; no total is shown when output pricing is unknown.
| Model | Lowest-input-rate channel | Input / 1M | Output / 1M | Example cost (USD) | Channels | Selected price verified |
|---|---|---|---|---|---|---|
| Claude Haiku 5.5 Batch | OpenRouter | $0.05 | $0.25 | $0.1125 | 1 | Oct 17, 2026 |
| Claude Haiku 5.5 | OpenRouter | $0.10 | $0.50 | $0.225 | 1 | Oct 17, 2026 |
| Claude Sonnet 5.5 Batch | OpenRouter | $1.00 | $5.00 | $2.25 | 1 | Oct 17, 2026 |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $10.00 | $4.50 | 2 | Oct 17, 2026 |
| Claude Opus 5.5 Batch | OpenRouter | $2.00 | $10.00 | $4.50 | 1 | Oct 17, 2026 |
| Claude Opus 5.5 | Anthropic | $4.00 | $20.00 | $9.00 | 2 | Oct 17, 2026 |
| Claude Fable 5.1 Batch | OpenRouter | $5.00 | $25.00 | $11.25 | 1 | Oct 17, 2026 |
| Claude Fable 5.1 | Anthropic | $10.00 | $50.00 | $22.50 | 2 | Oct 17, 2026 |
| Claude Opus 5 Batch | OpenRouter | $2.50 | $12.50 | $5.625 | 1 | Oct 17, 2026 |
| Claude Opus 5 | Anthropic | $5.00 | $25.00 | $11.25 | 2 | Oct 17, 2026 |
| Claude Sonnet 5 Batch | OpenRouter | $1.00 | $5.00 | $2.25 | 1 | Oct 17, 2026 |
| Claude Sonnet 5 | Anthropic | $2.00 | $10.00 | $4.50 | 2 | Oct 17, 2026 |
API cost versus subscription price, and included API credits
Max and Team plans include monthly Claude API credits: Max 5x $100, Max 20x $200; Team $20 per Standard seat and $100 per Premium seat, pooled across the team and capped at $500. Free, Pro and Enterprise get none. Credits are claimed by linking a Claude Console organization, after 7 days on an eligible plan. They cover only the Claude API, Claude Managed Agents, the Claude Agent SDK and the playground in the Claude Console, not Claude Code, extra usage in the Claude apps, or Claude on AWS, Bedrock, Google Cloud (Vertex AI) or Microsoft Foundry. They refresh each billing cycle (monthly on annual plans) and expire unused, with no rollover. Estimate API spend from token volume; anything above the credit is billed at normal API rates.
Cloud and aggregator channels
AWS Bedrock, Google Vertex AI, and aggregators may expose the same Claude family with different regional availability, contracts, rate limits, and billing. Price is one input, not proof that two channels are operationally identical.
Cost formula: cache writes and reads are different
Claude cost = (ordinary input × I + cache writes × W + cache reads × R + output × O) ÷ 1M + tool charges. I, W, R and O are the current rates for the exact model, cache lifetime and context tier. Writes use a premium over base input; reads use a lower multiplier, which can vary by model. Count each input category once. Eligible asynchronous Batch API input and output receive a 50% discount; choose the batch rate entries rather than applying another discount to rates already stored as batch prices.
Usage example: reuse a cached document
Assume 100 requests within one cache lifetime, each with the same 20,000-token document, 1,000 new input tokens and 500 output tokens, below the selected model’s context-tier threshold. Assume one successful cache write followed by 99 hits, with no tools or batch processing. The illustration costs 0.1 × I + 0.02 × W + 1.98 × R + 0.05 × O. Without caching it is 2.1 × I + 0.05 × O. Compare using the live rates; misses or expiry create more writes or ordinary input and change the result.
Easy-to-miss costs: context tiers and cloud regions
Check whether the selected model has a long-context tier; do not assume every Claude version does. Cache reads and writes can count toward the prompt-length threshold. Cache lifetime changes write pricing. Bedrock and Vertex AI regional or multi-region endpoints can have different rates from global routing; use the cloud provider’s rate card and invoice. Server-side tools add charges beyond the token estimate.
How to use this data
- Recalculate with your real input/output token mix.
- Check region, rate limits, and payment method.
- Recheck official sources before purchasing or switching.
Frequently asked questions
Can Max or Team API credits pay for Claude Code?
No. The monthly credits cover only the Claude API, Managed Agents, the Agent SDK and the playground in the Claude Console. Claude Code, extra usage in the Claude apps, and Claude on Bedrock, Vertex AI or Foundry are excluded, and unused credits expire each billing cycle.
Does every cached token use the read rate?
No. A cache creation uses the write rate for its lifetime; later hits use the read rate. Use the separate usage counters and model-specific multipliers.
Can I use the batch discount for interactive calls?
The 50% input/output discount applies to eligible asynchronous Batch API usage. A standard interactive request uses its standard rates.
Does long context always increase Claude rates?
It depends on the model. Check its current threshold and rate entries, including whether cached input contributes to prompt length.
Are Bedrock and Vertex AI rates identical to direct API rates?
Check the exact region, endpoint type and cloud contract. Regional pricing and billing terms can differ from the direct Claude API.
Official sources and review date
Billing guidance reviewed: . Price verification dates are listed separately in the table.