What $1,000 buys on each Claude model: Fable, Opus, Sonnet and Haiku

02The numbers (API prices, 20 Oct 2026)

1. Fable 5.1: $10 / $50 per MTok. $0.045 a message. 22,222 messages per $1,000, or 44,444 with the Batch API.
2. Opus 5.5: $4 / $20 per MTok. $0.018 a message. 55,555 messages per $1,000, or 111,111 with the Batch API.
3. Sonnet 5.5: $2 / $10 per MTok. $0.009 a message. 111,111 messages per $1,000, or 222,222 with the Batch API.
4. Haiku 5.5 (prompt up to 100K tokens): $0.10 / $0.50 per MTok. $0.00045 a message. 2,222,222 messages per $1,000, or 4,444,444 with the Batch API.

Worked example for Opus 5.5: 2,000 × $4 ÷ 1,000,000 = $0.008 for the input, and 500 × $20 ÷ 1,000,000 = $0.010 for the output. That makes $0.018 a message, and $1,000 ÷ $0.018 = 55,555 messages.

Haiku 5.5 gets 40 times more messages than Opus 5.5 for the same money, and 100 times more than Fable 5.1.

03Why it matters

Cost per message is not the only thing that matters. Anthropic positions Fable 5.1 for demanding reasoning and long-horizon agentic work. It suggests Opus 5.5 as the starting point for most workloads, Sonnet 5.5 as the balance of speed and intelligence, and Haiku 5.5 for high-volume work such as classification, extraction and routing. A common pattern is to send the bulk, simple calls to Haiku and only the hard ones to Opus or Fable. That way most of your budget goes 40 to 100 times further.

04Step by step: halve the bill with the Batch API

If you don't need the answer right away (evals, tagging, summarising a backlog), the Message Batches API is 50% off both input and output tokens on all four models. Most batches finish within an hour, and results are available within 24 hours.

1. Get an API key in the Claude Console (platform.claude.com) and set it in your shell: export ANTHROPIC_API_KEY=your-key
2. Send a batch. Each request has its own custom_id and normal Messages parameters (example below).
3. Poll the batch with the id it returns (GET /v1/messages/batches/<id>) until processing_status is ended, then download the results from its results_url.

curl https://api.anthropic.com/v1/messages/batches \
  --header "x-api-key: $ANTHROPIC_API_KEY" \
  --header "anthropic-version: 2023-06-01" \
  --header "content-type: application/json" \
  --data '{"requests": [
    {"custom_id": "ticket-1", "params": {"model": "claude-haiku-5-5", "max_tokens": 500,
      "messages": [{"role": "user", "content": "Tag this support ticket: ..."}]}}
  ]}'

Model IDs: claude-fable-5-1, claude-opus-5-5, claude-sonnet-5-5, claude-haiku-5-5.

05Good to know

1. Your messages are probably not 2,000 + 500. In a chat, every turn resends the history, so input grows each turn. Multiply by your real sizes. The Console and the token counting endpoint will tell you.
2. Prompt caching cuts input further. Cache reads cost a small fraction of the base input price, so a long, repeated system prompt gets much cheaper.
3. Extras that raise the price: US-only inference is 1.1x, and fast mode on Opus 5.5 is 2x standard pricing.
4. Claude.ai plans are different. Pro and Max subscriptions are not billed per token. These numbers are for the API.
5. Prices were checked on 20 Oct 2026 and can change. Check the pricing pages before you budget.

06Official links

• Claude pricing (API tab)
• Full pricing table, including Haiku's tiers and batch prices
• Batch processing guide
• Models overview and IDs