English · 6 min read
Claude Haiku 5.5: price, the per-task catch, and who should use it
Claude Haiku 5.5 launched on 7 October 2026 at $0.10 / $0.50 per 1M tokens, the same list price as GPT-6 Luna. An independent test found it uses about 3x Luna's output per task. What that means for your bill, with sources.
By Dapols ·
The short answer: Claude Haiku 5.5 is Anthropic's cheapest and fastest model, released on 7 October 2026. It costs $0.10 per 1M input tokens and $0.50 per 1M output tokens for prompts up to 100,000 tokens, and $0.50 / $2.50 for longer prompts (Anthropic pricing). That is a tenth of Haiku 4.5's $1 / $5, and the same list price as OpenAI's GPT-6 Luna. The catch: Artificial Analysis found it writes about three times as many output tokens per task as Luna at max effort, so a finished job can cost about three times more (Artificial Analysis).
Checked 8 October 2026. Prices and specs come from Anthropic's own docs; benchmark figures are labelled with who reported them.
What Claude Haiku 5.5 costs
All prices are per 1M tokens, from Anthropic's pricing page:
| Prompt up to 100K tokens | Prompt over 100K tokens | |
|---|---|---|
| Input | $0.10 | $0.50 |
| Output | $0.50 | $2.50 |
| Cache read | $0.01 | $0.05 |
| 5-minute cache write | $0.125 | $0.625 |
| Batch API (input / output) | $0.05 / $0.25 | $0.25 / $1.25 |
Two things are new here. First, Haiku 5.5 is the first Claude model priced by prompt length: once a prompt goes over 100,000 tokens, every token in that request costs five times more. Every other current Claude model bills its full 1M window at one rate. Second, the model id is simply claude-haiku-5-5, and it is on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry (Anthropic models overview).
The specs that matter
- Context window: 1M tokens, with up to 128K output tokens per request.
- Effort settings: a first for Haiku. The default is medium; you can raise or lower it. Higher effort thinks longer and writes more.
- What Anthropic says it is for: "high-volume, latency-sensitive tasks such as classification, extraction, and routing", and it is listed as the fastest model in the current Claude lineup.
How good is it?
Anthropic's own figures, as reported by VentureBeat:
- Terminal-Bench 4.0 (agentic coding): 39.2% at max effort, against GPT-6 Luna's 16.4% and Haiku 4.5's 0.0%. At the default medium effort, VentureBeat reports about 20%.
- OSWorld 2.1, offline subset (operating a computer): 72.4%, against Luna's 48.9% and Haiku 4.5's 15.7%.
These are vendor numbers. The independent check is Artificial Analysis, which scores Haiku 5.5 at 43 on its Intelligence Index at max effort, up 26 points on the last Haiku, and ahead of GPT-6 Luna.
The per-task catch
A low price per token only helps if the model does not use many more tokens. Artificial Analysis measured how much each model writes to finish one task in its index:
| Model and effort | Intelligence Index | Output tokens per task |
|---|---|---|
| Claude Haiku 5.5, max | 43 | about 162,000 |
| Claude Haiku 5.5, high | 38 | about 55,000 |
| GPT-6 Luna, max | 38 | about 50,000 |
Source: Artificial Analysis, 7 October 2026.
At max effort, Haiku 5.5 writes roughly three times what Luna writes. Because both bill $0.50 per 1M output tokens, the job costs roughly three times more. At high effort the gap mostly closes: Haiku reaches Luna's score with a similar number of tokens. Artificial Analysis also notes its cost figures do not yet model the over-100K price step, so treat any published "cost per task" as provisional.
What to do with that: start at low or medium effort, measure the real bill on your own task for a day, and only raise effort where quality actually fails.
Haiku 5.5 vs GPT-6 Luna vs DeepSeek V4.1 Flash
| Claude Haiku 5.5 | GPT-6 Luna | DeepSeek V4.1 Flash | |
|---|---|---|---|
| Price per 1M (in / out) | $0.10 / $0.50 up to 100K-token prompts | $0.10 / $0.50 | $0.15 / $0.60 off-peak, double at peak |
| Long prompts | $0.50 / $2.50 over 100K | $0.20 / $0.75 over 272K input tokens | Priced by time of day, not prompt length |
| Open weights | No | No | Yes (MIT) |
Sources: Anthropic, our model tracker for Luna and DeepSeek, checked 8 October 2026.
Who should use Claude Haiku 5.5
- Use it for short, high-volume jobs where you already like Claude's writing: sorting incoming email, pulling fields out of invoices, tagging support tickets, routing requests to the right person. Short prompts stay in the cheap tier.
- Test it against Luna if you run agents that work on their own for many steps. Haiku scores higher, but the token count decides the bill.
- Skip it for long documents. A prompt over 100K tokens costs five times more, so a whole contract or a long chat history is cheaper on a model with one flat rate.
- Skip it for your hardest coding work. Claude Sonnet 5.5 and Opus 5.5 are far ahead on coding benchmarks; see best AI model for coding.
For a small business, the honest summary is that this is a good cheap model for repetitive tasks, not a reason to switch everything. If you are not sure which jobs in your business are worth automating in the first place, that is what the free 2-minute AI plan finder answers.
Frequently asked questions
When was Claude Haiku 5.5 released? On 7 October 2026, on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry.
How much does Claude Haiku 5.5 cost? $0.10 per 1M input tokens and $0.50 per 1M output tokens for prompts up to 100,000 tokens, and $0.50 / $2.50 for longer prompts. Cache reads are $0.01 per 1M in the lower tier.
Is Claude Haiku 5.5 cheaper than GPT-6 Luna? Not per token: both list at $0.10 / $0.50. Per task it can be more expensive, because Artificial Analysis found Haiku 5.5 at max effort uses about 3x Luna's output tokens.
What is the Claude Haiku 5.5 model id?
claude-haiku-5-5, on the Claude API and on the cloud platforms.
Is Claude Haiku 5.5 better than Haiku 4.5? Yes, on every benchmark Anthropic published, and at a tenth of the price for prompts under 100K tokens. Haiku 4.5 is now a legacy model, still available.