Claude API pricing in 2026: cost per million tokens

How much does each Claude model cost per million tokens?

Anthropic prices every model per million tokens (MTok), with separate rates for input, output, cache writes and cache hits. The four current models lead the table. Older models stay on sale at their own rates. Figures come from Anthropic’s pricing page. Prices checked November 10, 2026.

Claude Mythos 5.2 lists the same rates as Fable 5.2. Mythos 5.1 and Mythos 5 match Fable 5.1 and Fable 5. Opus 4.5, 4.6 and 4.7 share the Opus 4.8 rates, and Sonnet 4.5 and Sonnet 4 share the Sonnet 4.6 rates.

How do batch, caching and other options change the price?

The list rate is the starting point. Several options move it up or down, and Anthropic says they stack. Batch and caching together give the lowest price per token.

• Batch API: 50% off input and output for asynchronous jobs. Sonnet 5.5 drops to $1 input and $5 output.
• Cache hits: 0.1x the input rate on most models, 0.05x on Opus 5.5 and Sonnet 5.5, and 0.025x on Fable 5.2, Fable 5.1, Mythos 5.2 and Mythos 5.1.
• Cache writes: 1.25x the input rate for a 5-minute cache and 2x for a 1-hour cache. A 5-minute cache pays off after one hit.
• Long context: Claude 4.6 and later models, except Haiku 5.5, bill the full 1 million token window at standard rates. Haiku 5.5 bills a whole request at 5x once its prompt passes 100,000 tokens.
• US-only inference: setting inference_geo to "us" adds 10% to every token on Claude 4.6 and later.
• Fast mode: $16 input and $80 output on Fable 5.2, $8 and $40 on Opus 5.5, and $10 and $50 on Opus 5 and Opus 4.8. It does not work with the Batch API.
• Server tools: web search costs $10 per 1,000 searches plus tokens. Web fetch adds only tokens. Code execution includes 1,550 free hours a month per organization, then $0.05 per container hour.

What does a real Claude workload cost?

Take 10,000 calls, each with a 2,000-token prompt and a 500-token answer. That is 20 million input tokens and 5 million output tokens. The cached column assumes 1,500 tokens of each prompt are a cache hit, such as a fixed system prompt, and ignores the small cost of cache writes.

Two things raise real bills. Claude 4.7 and later models use a new tokenizer that produces about 30% more tokens for the same text. The Locus catalog also notes that extended thinking tokens bill at the output rate, so a model that thinks before it answers can cost several times the table figure.

Is there a free tier, and what are the rate limits?

There is no standing free tier. Anthropic gives new users a small amount of free credit to test the API, then bills monthly usage by card or invoice.

Limits depend on your usage tier. New organizations can start in an Evaluation tier with lower limits that rise as account history builds. Each tier also has a monthly spend cap. When you hit it, the API returns errors until 00:00 UTC on the first of the next month unless you get a higher limit. Cached tokens do not count toward input token limits.

How is Claude billed through Locus Pro?

Locus Pro offers Claude as the anthropic/chat endpoint, billed from one prepaid credit balance that also covers other providers, with no Anthropic account, key or minimum. The route accepts Anthropic’s own Messages request body. Catalog checked November 10, 2026.

Each call places a hold sized from your prompt and max_tokens, then settles to the tokens Anthropic reports. You never pay more than the hold. Send max_credits to refuse any call whose hold would exceed your ceiling.

• Claude pay per use through Locus Pro
• Browse model endpoints in the catalog

Which Claude option should you choose?

Open an Anthropic account when Claude is your main model and volume is steady. You get the Batch API and the newest models first, and Claude can cost less per call with your own Anthropic account and API key.

Locus Pro fits an agent that calls Claude next to search, scraping or data APIs, with low or uneven volume. You keep one bill, set a price ceiling per call and skip a separate Anthropic account. The Free plan allows 5 tool calls a minute, and Enterprise lets you bill your own customers for their agents’ calls.

• Classification, extraction and routing: Claude Haiku 5.5, the cheapest model at $0.10 and $0.50 for prompts under 100,000 tokens.
• Most production work: Claude Sonnet 5.5 at $2 and $10.
• Long agentic coding and knowledge work: Claude Opus 5.5 at $4 and $20.
• The hardest reasoning: Claude Fable 5.2 at $8 and $40.
• Large jobs that can wait: any model through the Batch API on an Anthropic account.

• Cheapest way to call AI models by API
• Locus Pro plans