On October 7, 2026, Anthropic launched Claude Haiku 5.5 — the third model in its Claude 5.5 family in the past month, ahead of a planned IPO. The headline number: $0.10 per million input tokens and $0.50 per million output tokens for short prompts, down from $1 and $5 on Haiku 4.5. That’s a 90% price cut on the sticker. For small businesses running AI automations, it could slash monthly API bills from the cost of a nice dinner to the cost of a coffee. But there’s a catch in the fine print — the cheap rate only applies to prompts under 100,000 tokens, and the price jumps fivefold the moment you cross that line.
Here’s what the launch actually means for your automation budget, and whether you should care.
What Claude Haiku 5.5 is (and who it’s for)
Haiku has always been Anthropic’s small, fast, cheap model — the one you reach for when you need to run the same narrow job thousands of times a day. Anthropic positions Haiku 5.5 for exactly that: classification, summarization, extraction, live customer support, voice agents, and in-app assistants. It can also run as a subagent alongside the bigger Sonnet 5.5 or Opus 5.5 models, handling high-volume grunt work while the heavy model does the architecture.
The new model isn’t just cheaper. It got a bigger brain too: context window jumped from 200K to 1 million tokens, max output doubled to 128K tokens, and it’s the first Haiku with an adjustable effort setting, letting developers trade latency and cost against how hard the model thinks. Early benchmarks look strong for a small model: 72.4% on the offline subset of OSWorld 2.1 (up from 15.7% on Haiku 4.5), 39.2% on Terminal-Bench 4.0 where its predecessor scored zero, and 46.4% on FrontierCode 1.1, according to testing-catalog’s roundup of the launch.
It’s also the first Haiku with built-in safeguards for a narrow set of high-risk cybersecurity requests, though Anthropic says everyday tasks won’t be affected (see Anthropic’s official announcement). Availability is broad from day one: the Claude API under model ID claude-haiku-5-5, the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Foundry.
For a deeper look at the rest of Anthropic’s and OpenAI’s current lineup, see our AI tools for small business owners guide — and if you’re weighing vendors, OpenAI’s budget option is covered in our GPT-6 Sol and Luna rundown.
The new pricing, line by line
The price list has two tiers, and which one you pay depends entirely on prompt length:
Prompts up to 100,000 tokens:
– Input: $0.10 per million tokens
– Output: $0.50 per million tokens
– Cache reads: $0.01 per million tokens
– Cache writes: $0.125 (5-minute) or $0.20 (1-hour) per million tokens
Prompts over 100,000 tokens:
– Input: $0.50 per million tokens
– Output: $2.50 per million tokens
– Cache reads: $0.05 per million tokens
– Cache writes: $0.625 / $1.00 per million tokens
Compared to Haiku 4.5’s flat $1/$5, the low tier is a 90% cut and the high tier is still 50% cheaper. Anthropic estimates average running costs land about 75% below Haiku 4.5 — not 90% — because the new tokenizer uses somewhat more tokens for equivalent work and some requests land in the higher tier. Reuters reported the same 75% figure at launch.
The 100,000-token catch nobody reads
Here’s the number that matters most: per a dev.to analysis of the release, a request with a 100,000-token prompt and 2,000 tokens of output costs $0.011. Add one token to the prompt and the same request costs $0.055. Nothing else changed — you just crossed a line in the price list.
The good news is that roughly 90% of requests to Haiku 4.5 fell within the cheaper tier, per Anthropic’s own figures cited by testing-catalog. Small-business workloads — classifying support tickets, summarizing reviews, extracting invoice data — almost always run well under 100K tokens per request. A 100,000-token prompt is roughly 75,000 words of context, or a few hundred pages. If your automation sends that much context per call, you’re doing something unusual (or your agent framework is stuffing the whole conversation history into every request).
The practical takeaway: watch your prompt sizes. If you build retrieval or agent workflows, keep context windows lean — not just for speed, but because crossing the threshold quintuples your bill for that request.
What this means for your monthly AI bill
Let’s do real math with the published prices. Say you run a support bot that handles 10,000 customer messages a month (a healthy volume for a small business). Assume each exchange uses 500 input tokens of context and 150 tokens of output:
- Input: 10,000 × 500 = 5 million tokens → 5 × $0.10 = $0.50
- Output: 10,000 × 150 = 1.5 million tokens → 1.5 × $0.50 = $0.75
- Total: $1.25/month
On Haiku 4.5, the same workload cost $5 in input plus $7.50 in output — $12.50 a month. On a per-token basis, that’s a genuine 90% cut.
Now scale it up. An agency running document summarization — say 2,000 invoices a month, each needing 2,000 input tokens and 100 output tokens:
- Input: 4 million tokens → $0.40
- Output: 200,000 tokens → $0.10
- Total: $0.50/month
These numbers are so small they change the decision calculus entirely. Tasks you might have run weekly to save money can now run in real time. Anthropic’s own 75%-average-savings estimate is the more conservative figure to budget against (it accounts for tokenizer changes and tier mix), but either way we’re talking dollars per month, not hundreds.
One honest caveat from the fine print analysis: token prices alone don’t set your bill. Retries, accuracy (a cheap model that needs two attempts costs double), and cache usage matter. Still, Haiku 5.5’s pricing matches OpenAI’s GPT-6 Luna at $0.10/$0.50, which means the two cheapest frontier-vendor options are now priced identically at the low tier — the differentiator is how few tokens each needs to get the job right.
Where you can actually use it
Haiku 5.5 is live now — no waitlist. You can call it through:
- The Claude API with model ID
claude-haiku-5-5 - The Claude Platform and Claude.ai
- Amazon Bedrock (
anthropic.claude-haiku-5-5), Google Cloud, and Microsoft Foundry
If you already run Haiku 4.5 workloads, check the migration guide before switching — the new pricing tiers and the effort setting change how you tune requests.
Should your business switch?
Switch (or start) with Haiku 5.5 if:
– You run high-volume, narrow tasks: support triage, review summaries, data extraction, lead qualification
– Your prompts stay under 100,000 tokens (they almost certainly do)
– You want a subagent for grunt work under a bigger model’s direction
Stay where you are if:
– Your current setup is locked into another vendor’s ecosystem and migration would cost real engineering time — the savings are real but small in absolute dollars
– You need the heaviest reasoning or agentic coding — Anthropic still positions Sonnet 5.5 and Opus 5.5 for that
The bottom line: this launch isn’t really about Anthropic versus OpenAI’s pricing war. It’s about the floor dropping out of automation costs for small businesses. When a month of AI support costs less than one coffee, “should we automate this?” stops being a budget question and becomes a build question.
Frequently Asked Questions
When was Claude Haiku 5.5 released?
October 7, 2026 — the third Claude 5.5 model Anthropic released in about a month, ahead of its planned IPO.
How much does Claude Haiku 5.5 cost?
$0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, with cache reads at $0.01 per million. Above 100,000 tokens, rates rise to $0.50/$2.50 per million — a fivefold jump.
How much cheaper is Haiku 5.5 than Haiku 4.5?
90% cheaper on the sticker price for short prompts ($0.10 vs $1.00 input). Anthropic estimates average real-world workloads cost about 75% less, accounting for the new tokenizer and pricing tiers.
What is the context window of Claude Haiku 5.5?
1 million tokens of context with up to 128K output tokens — up from 200K/64K on Haiku 4.5.
How does Haiku 5.5 compare to GPT-6 Luna on price?
They match: both are $0.10 per million input tokens and $0.50 per million output at their base tiers, per a VentureBeat comparison of official pricing checked October 7.
Does Haiku 5.5 have any new safety features?
Yes — it’s the first Haiku with built-in safeguards for a narrow set of high-risk cybersecurity requests. Anthropic says most everyday tasks are unaffected.