Claude API Pricing (2026)
This page lists Anthropic's current Claude API prices, explains the two discount mechanisms that change your effective rate, and covers a tokenizer detail that makes Claude bills differ across model generations even at the same list price.
Claude model prices
List prices verified 2026-10-11. USD per 1 million tokens, standard tier.
| Model | Input $/1M | Cache read $/1M | Output $/1M | Context | Max output |
|---|---|---|---|---|---|
| Claude Fable 5.1 | $10.00 | $0.25 | $50.00 | 1,000,000 | 128,000 |
| Claude Opus 5.5 | $4.00 | $0.20 | $20.00 | 1,000,000 | 128,000 |
| Claude Sonnet 5.5 | $2.00 | $0.10 | $10.00 | 1,000,000 | 128,000 |
| Claude Haiku 5.5 (up to 100K prompt) | $0.10 | $0.01 | $0.50 | 1,000,000 | 128,000 |
| Claude Haiku 5.5 (over 100K prompt) | $0.50 | $0.05 | $2.50 | 1,000,000 | 128,000 |
| Claude Fable 5 (legacy) | $10.00 | $1.00 | $50.00 | 1,000,000 | 128,000 |
| Claude Opus 5 (legacy) | $5.00 | $0.50 | $25.00 | 1,000,000 | 128,000 |
| Claude Opus 4.8 (legacy) | $5.00 | $0.50 | $25.00 | 1,000,000 | 128,000 |
| Claude Sonnet 5 (legacy) | $2.00 | $0.20 | $10.00 | 1,000,000 | 128,000 |
| Claude Sonnet 4.6 (legacy) | $3.00 | $0.30 | $15.00 | 1,000,000 | 128,000 |
| Claude Haiku 4.5 (legacy) | $1.00 | $0.10 | $5.00 | 200,000 | 64,000 |
Prompt cache pricing
Prompt caching changes the input rate on repeated content. The mechanics:
- Cache reads bill at 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. On Sonnet 5.5 that is $0.10 per 1M tokens instead of $2.00.
- Cache writes bill at 1.25x the input price for the 5-minute TTL, or 2x for the 1-hour TTL.
The write premium is small next to the read discount. Content that gets written once and read many times, a system prompt, tool definitions, a document under discussion, costs 90% to 97.5% less on every read after the first request, depending on the model. For agent and coding workloads where the same prefix repeats on every step, cached reads become most of the input bill.
Batch API: 50% off
The Batch API applies a 50% discount to both input and output. It suits any job that does not need an immediate response, such as bulk classification, evaluation runs, and offline content generation.
The tokenizer difference between generations
Claude has no public tokenizer. Exact counts come from the free count_tokens API. That matters here because the tokenizer differs by model generation. Opus 4.7 and later, including every 5.x model, use a newer tokenizer that produces roughly 1x to 1.35x the tokens of the 4.6-and-older family on the same text. Sonnet 5 counts about 30% more tokens than Sonnet 4.6 for identical text.
So the same prompt can bill differently across Claude generations for two separate reasons: the price per token, and the number of tokens the prompt becomes. Sonnet 5 lists at $2.00 per 1M input tokens against $3.00 on Sonnet 4.6, but the same text counts as about 30% more tokens on Sonnet 5, so the real saving is smaller than the list prices suggest. When you compare Claude models, count the prompt with the count_tokens API for each model rather than assuming one count.
Related pages
- Claude Code cost per month, estimating API-rate spend for coding sessions.
- Prompt caching guide, how to structure prompts so the cache actually hits.
- Claude context windows and the Claude token counter.
- Compare against Gemini pricing, OpenAI pricing, or the all-provider table.