Claude Haiku 5.5 review
Anthropic's small, fast model, 90% cheaper per token than Haiku 4.5.
A huge price cut for a small model: 90% cheaper per token than Haiku 4.5, with far better scores. It beats GPT-6 Luna on every test where both were measured. A strong default for high-volume jobs.
What is Claude Haiku 5.5?
Claude Haiku 5.5 is Anthropic's small, fast model, released on 7 October 2026. It completes the Claude 5.5 family after Opus 5.5 and Sonnet 5.5, and is built for high-volume, low-cost work like summaries, helper agents and browser tasks.
Key features
- Costs $0.10 per 1M input tokens, 90% less than Haiku 4.5
- First Haiku model with an adjustable thinking effort
- New tools for using a browser and a computer
- Reads 1 million tokens at once and writes up to 128,000
- Big jump in skill: 72.4% on OSWorld 2.1, against 15.7% for Haiku 4.5
Specs and price
| Made by | Anthropic |
| Released | 7 Oct 2026 |
| API price | $0.10 in / $0.50 out per 1M tokens, for prompts up to 100K tokens |
| Longer prompts | $0.50 in / $2.50 out per 1M tokens |
| Batch | 50% off |
| Real-world saving | About 75% on average (Anthropic), because the new tokenizer counts about 30% more tokens for the same text |
| How much it can read | 1,000,000 tokens |
| Longest answer | 128,000 tokens |
| Input | Text and images |
| Where | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Free to download | No |
| Name in the API | claude-haiku-5-5 |
Test scores
| Test | Score | Published by |
|---|---|---|
| OSWorld 2.1Using a computer like a person | 72.4% (GPT-6 Luna: 48.9%) | Anthropic |
| Terminal-Bench 4.0Coding agents working in a terminal | 39.2% (GPT-6 Luna: 16.4%) | Anthropic |
| GDPval-AA v2.1Real office work (rating) | 1620 (GPT-6 Luna: 1437) | Anthropic |
| Humanity's Last ExamVery hard expert questions, with tools | 57.4% (Haiku 4.5: 18.7%) | Anthropic |
| Artificial Analysis indexOverall score, high effort | 37.8 | Independent |
Scores published by Anthropic are the maker's own numbers. We will add our own results in the hands-on review.
How to use Claude Haiku 5.5
- For developers: call
claude-haiku-5-5in the Claude API, or use it through Amazon Bedrock, Google Cloud or Microsoft Foundry. - Pick an effort level, from low to max (medium by default), to trade quality for speed and cost.
- Send jobs that can wait through the Batch API for another 50% off.
Who it is for
Good fit
- Apps that send lots of requests: support chat, summaries, tagging
- Helper agents inside bigger agent systems
- Browser and computer automation on a budget
Skip it if
- The hardest reasoning or coding jobs: Sonnet 5.5 and Opus 5.5 score clearly higher
- Apps that set temperature or prefill answers: those now return errors
Pros and cons
Pros
- Very low price per token
- Beats GPT-6 Luna wherever both were tested
- 1M token memory
- Thinking effort you can tune
Cons
- The new tokenizer counts about 30% more tokens, so real savings are about 75%, not 90%
- Prompts over 100K tokens cost five times more
- Several older API settings now return errors
- Trails Claude Sonnet 5.5 on every published test
How it compares
Against GPT-6 Luna, OpenAI's small model, Haiku 5.5 scores higher on every test where both were reported, for example computer use (72.4% against 48.9%). For heavier work, Claude Sonnet 5.5 scores higher on everything but costs more.
Hands-on review In testing
We are running Claude Haiku 5.5 on real tasks. Results will be added to this page, at the same address.
How it got here
| Date | What happened | Source type |
|---|---|---|
| Oct 7, 2026 | Claude Haiku 5.5 released at $0.10 / $0.50, completing the Claude 5.5 family.DataCamp, Claude Haiku 5.5 explained | Confirmed |
Frequently asked questions
When was Claude Haiku 5.5 released?
On 7 October 2026, two weeks after Claude Opus 5.5.
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. Longer prompts cost $0.50 and $2.50.
Is Claude Haiku 5.5 free?
No. It cannot be downloaded, and the API is paid per token. It is one of the cheapest capable models available.
Is Claude Haiku 5.5 better than Haiku 4.5?
Much better on every test Anthropic published, for example 72.4% against 15.7% on computer use, and 90% cheaper per token.
Claude Haiku 5.5 or GPT-6 Luna?
Haiku 5.5 scored higher on every test where both were measured, including computer use, coding agents and office work.
Related models
Claude Opus 5.5
AnthropicAnthropic's top model, cheaper and faster than Opus 5.
Claude Fable 5.2
AnthropicPrediction markets expect a Fable point release. Anthropic has not announced anything.
Claude 6
AnthropicThe next major Claude generation. Only prediction markets give a date so far.