Claude Batch API for SEO: half-price bulk jobs and what they cost
Batch rates per model, the meta description and keyword jobs that fit, the 24-hour and 100,000-request limits, and how caching cuts the bill further.
By ClaudeGPT Si Newsroom, Editorial team··3 min read
The Claude Batch API runs large sets of requests in the background at half the usual token price, which fits SEO work that does not need an instant reply, such as drafting meta descriptions for a whole catalog or sorting a keyword list by intent. As of November 2026, Sonnet 5.5 costs $1 per million input tokens and $5 per million output tokens in a batch, against $2 and $10 for live calls.
Figures here come from Anthropic's batch processing documentation and its pricing page, checked on November 4, 2026.
Batch prices next to live prices
| Model | Live input / output | Batch input / output |
|---|---|---|
| Claude Fable 5.1 | $10 / $50 | $5 / $25 |
| Claude Opus 5.5 | $4 / $20 | $2 / $10 |
| Claude Sonnet 5.5 | $2 / $10 | $1 / $5 |
| Claude Haiku 5.5, prompts up to 100K tokens | $0.10 / $0.50 | $0.05 / $0.25 |
| Claude Haiku 4.5 | $1 / $5 | $0.50 / $2.50 |
All prices are US dollars per million tokens. The discount applies to both input and output.
SEO jobs that suit a batch
Anything you would otherwise loop over one page or one keyword at a time is a candidate.
- Title tags and meta descriptions for thousands of product or category pages.
- Labelling a keyword export as informational, commercial or navigational.
- Short answer blocks or FAQ drafts built from text you already have on the page.
- Checking existing copy against a brand or style rule and flagging misses.
Anthropic positions Haiku 5.5 for high-volume classification and extraction, which covers the labelling jobs, while Sonnet 5.5 is the safer choice for copy a person will read.
A rough example, with assumed sizes: 10,000 meta descriptions, each request sending 1,000 tokens of page text and instructions and getting 60 tokens back. On Sonnet 5.5 that is 10 million input and 600,000 output tokens, about $26 live or $13 as a batch. On Haiku 5.5 the batch comes to roughly $0.65. Your real token counts will differ, so run a small sample first.
Limits and timing
- A batch holds up to 100,000 requests or 256 MB, whichever comes first.
- Most batches finish within an hour. Results become available when every request is done, or after 24 hours.
- Requests still unfinished after 24 hours expire, and expired requests are not billed.
- Results stay downloadable for 29 days.
- Results can come back in any order, so give each request a
custom_id, such as the page URL or ID, and match on that. - Opus 5.5 fast mode does not work in batches.
Stack it with prompt caching
The batch discount and caching discounts combine. Put your shared instructions and examples at the start of every request so they form the same prefix, and the repeated part is billed at the cache read rate. Anthropic says cache hits inside batches are best effort, typically between 30 and 98 percent, and suggests the 1-hour cache lifetime for context that every request shares. The prefix has to clear the model's minimum, which is 512 tokens on Sonnet 5.5 and Haiku 5.5.
Keeping track of the spend
For the batch jobs themselves, the usage page in the Claude Console and your monthly invoice are the record. Most SEO teams also write and debug these scripts with a coding assistant, and that side costs money too. UsageAtlas, a free and open-source desktop tracker, shows the tokens and estimated cost of Claude Code, Codex and other coding tools on your machine without reading your prompts. If you are still deciding how to pay for that coding work, its write-up on choosing between an API key and a plan walks through the comparison.