Claude Haiku 5.5 (2026): API Pricing, the 100K-Token Price Line, and What It Costs on Real Work

Small models don't get headlines. They get the bill. If your business runs anything through the Claude API at volume, tagging tickets, summarizing calls, pulling fields out of invoices, this is the launch that moves the number.

Here's what Anthropic shipped, what it costs as of October 8, 2026, two workloads priced out, and the one design rule the new price sheet quietly creates.

What is Claude Haiku 5.5?

It's the third model in the Claude 5.5 family, after Opus 5.5 (September 22) and Sonnet 5.5 (September 28). Anthropic calls it "the cheapest, fastest, and most capable small model we've ever released," aimed at "quick and repetitive workloads (like summaries, compactions, database queries, and classification requests)."

• Speed: Anthropic's fastest model at standard speed, which is why it's pitched for live customer support and browser use.
• Effort control: the first Haiku with an adjustable effort setting, so you can trade cost against quality per task.
• Where it runs: the Claude API (model ID claude-haiku-5-5), AWS, Google Cloud and Microsoft Azure, Claude Code, and the Claude app on Free, Pro, Max, Team and Enterprise.

Anthropic's own benchmark table puts it far ahead of Haiku 4.5: 72.4% vs 15.7% on OSWorld 2.1 computer use (offline subset), 1620 vs 735 on GDPval-AA v2.1 knowledge work. Sonnet 5.5 still scores higher on every row, and Anthropic says plainly that Sonnet and Opus "remain better choices for complex agentic coding tasks." Haiku is the fast hands. The bigger models are still the planner. If you're choosing a default model for a business, that call lives in the best Claude model for business.

Claude Haiku 5.5 pricing

Per million tokens, from Anthropic's pricing docs, read October 8, 2026:

The Batch API halves Haiku 5.5 again: $0.05 / $0.25 under 100K, $0.25 / $1.25 over. Anthropic's headline is "around 75% less to run" than Haiku 4.5 on average. That's lower than the 90% rate cut for two reasons it states itself: some requests land over 100K, and the new tokenizer "uses slightly more tokens per task." Every other model's rates, plus the subscription side, are in Anthropic API pricing.

What does it cost on real work?

Rate-card arithmetic, no caching, and no adjustment for the tokenizer, so treat these as close estimates, not quotes.

Job 1: support ticket triage, 20,000 tickets a month

Each ticket: about 1,500 tokens in (the ticket plus your routing rules) and 300 out (a category, a priority, a draft first reply). That's 30 million input tokens and 6 million output a month.

Six dollars. For most businesses, that job just stopped being a line item.

Job 2: questions against long documents, 1,000 a month

Each request sends a 150,000-token document (a long contract, a quarter of call transcripts) and gets about 1,000 tokens back. That's 150 million input tokens, and every one of those prompts is over the 100K line.

Same model, same text, five times the bill, purely because of where the prompt length lands. The chunked figure ignores overlap between chunks and the extra calls to merge answers, so the real saving is a bit smaller. The direction isn't.

The 100,000-token line is a design rule

Under 100K tokens, Haiku 5.5 is the cheapest Claude by a mile. Over it, it's still cheaper than Haiku 4.5, but only by half. So the build question changes from "which model?" to "how big is each request?"

• Keep standing instructions short. A 20-page system prompt on every call eats the budget you could spend on the actual document.
• Chunk long inputs on purpose. Split by section, ask per section, merge at the end. Haiku is fast enough that the extra calls don't hurt.
• Let a bigger model plan and Haiku do the reading. That's the subagent pattern Anthropic is pitching: Opus or Sonnet breaks down the job, Haiku runs the many small pieces.

Does this change your Claude plan?

Not the price. Pro, Max, Team and Enterprise cost what they did on October 7, and the full tier map is in Claude AI pricing. Two things did change for subscribers:

• Haiku 5.5 is in the model picker on every plan, Free included, for when you want speed over depth.
• Max and Team now come with a monthly API credit, announced the same day: $100 on Max 5x, $200 on Max 20x, and $20 per Standard Team seat ($100 per Premium) pooled up to $500 a month. A 10-person Team on Standard seats gets $200 a month, which covers Job 1 above more than thirty times over. Who qualifies, how to claim it and what it won't pay for are in Claude API credits.

Should your business switch to Haiku 5.5?

• Switch now if you run Haiku 4.5 on classification, extraction, summaries or routing. Same job shape, better model, roughly a tenth of the rate under 100K. Test on a few hundred of your real inputs first.
• Try it as a downgrade for anything you run on Sonnet that is narrow and repetitive. Many "we use Sonnet for everything" setups are paying 20 times the rate for work Haiku can do.
• Don't switch the planner in a multi-step agent, or complex coding work. Anthropic says the bigger models stay better there, and its own numbers back that up.

If the honest answer is "we don't run anything on the API, we just have a few people in the chat app," the launch matters less than the setup. Getting a team's Claude workspace connected to the tools it already uses is my Claude for Small Business install: from $4,500, about two weeks, roughly three hours of your time. Seats are paid to Anthropic.

Sources

• Anthropic, "Introducing Claude Haiku 5.5", October 7, 2026, anthropic.com/claude-haiku-5-5 (read October 8, 2026)
• Anthropic, Claude Haiku product page, anthropic.com/claude/haiku (read October 8, 2026)
• Anthropic, Pricing, platform.claude.com/docs/en/about-claude/pricing (read October 8, 2026)
• Anthropic Help Center, "Monthly API credits for Max and Team plans", support.claude.com/en/articles/17154008 (read October 8, 2026)