Claude Haiku 5.5 is Anthropic’s newest small model, and it costs just $0.10 per million input tokens. That’s a 90% cut from Haiku 4.5 on prompts up to 100K tokens. The catch: go past 100K and the price jumps fivefold.
Here are the facts you need before you touch your API settings.
- Launched October 7, 2026, with the model ID
claude-haiku-5-5 - Short prompts: $0.10 input and $0.50 output per million tokens
- Prompts over 100K tokens: $0.50 input and $2.50 output
- Available on the Claude Platform, AWS, Google Cloud and Azure
- Max and Team plans get monthly API credits starting this week
Skim that and you’ve got the story. The rest of this post covers the numbers behind it and who should switch.
What Claude Haiku 5.5 costs, and where the 100K rule bites
Anthropic lists two price tiers, split by prompt size. Below the line, Haiku 5.5 is far cheaper than its predecessor. Above it, the savings shrink to half.
| Charge (per 1M tokens) | Up to 100K | Over 100K | Haiku 4.5 |
|---|---|---|---|
| Input | $0.10 | $0.50 | $1.00 |
| Output | $0.50 | $2.50 | $5.00 |
| Cache reads | $0.01 | $0.05 | $0.10 |
| Cache writes | $0.125 | $0.625 | $1.25 |
According to VentureBeat, Anthropic says about 90% of Haiku 4.5 requests land below 100K tokens. So most apps will see the low rate most of the time.
Don’t expect a clean 90% drop on your bill, though. A new tokenizer uses somewhat more tokens for the same text, so Anthropic puts the typical saving nearer 75%.
How the new small model scores against Haiku 4.5 and GPT-6 Luna
Anthropic’s own benchmark table shows a huge jump over Haiku 4.5. These are vendor-reported scores, not independent tests, so treat them as a starting point.
| Benchmark | Haiku 5.5 | Haiku 4.5 | GPT-6 Luna | Sonnet 5.5 |
|---|---|---|---|---|
| OSWorld 2.1 (offline subset) | 72.4% | 15.7% | 48.9% | 83.9% |
| Terminal-Bench 4.0 | 39.2% | 0.0% | 16.4% | 70.6% |
| Humanity’s Last Exam (no tools) | 45.9% | 10.2% | n/a | 56.9% |
| FrontierCode 1.1 | 46.4% | n/a | 42.4% | 52.1% |
Sonnet 5.5 still wins every row. Haiku 5.5 beats GPT-6 Luna on the ones where both have scores, and Luna charges the same $0.10 and $0.50 base rates.
One caveat on coding. The 39.2% Terminal-Bench score is at maximum effort. VentureBeat reports the default medium setting lands around 20%, so test on your own tasks.
Who should switch, and who should stay on Sonnet
Anthropic pitches Haiku for summarizing, classifying and database queries. For hard coding work, it still points people to Opus and Sonnet.
That matches the numbers. If your app sends lots of short, repetitive requests, Haiku 5.5 is an easy win. If you need the strongest agent or coding results, Sonnet is still the pick, and our breakdown of Claude Sonnet 5.5 pricing and benchmarks shows what you get for $2 input and $10 output.
Sonnet users get a small bonus too. Its cache-read price drops from $0.20 to $0.10 per million tokens, which Anthropic says saves about 20% on typical agent workloads.
A quick checklist before you migrate to Claude Haiku 5.5
Swapping the model ID takes a minute. Avoiding a surprise invoice takes a little more care, so run through these checks first.
- Log your real prompt sizes, because the 100K line decides which price tier you pay.
- Compare token counts on the same text, since the updated tokenizer can use somewhat more tokens.
- Test at the default medium effort first, then raise it only for tasks that fail.
- Re-run your own evals, because the published scores come from Anthropic.
Why bother? Small price differences multiply fast at millions of requests a day, and a wrong tier guess can erase the savings you were chasing. Anthropic also added beta browser and computer operation to its Python and TypeScript SDKs with this release, which is worth a look if you build agents.
Free API credits for Max and Team subscribers
Anthropic is also adding monthly API credits for paid plans, rolling out this week. They work on any model on its platform.
| Plan | Monthly API credit |
|---|---|
| Max 5x | $100 |
| Max 20x | $200 |
| Team | Up to $500, shared across users |
The credits help if you build side projects on the API. If you mostly use Claude inside Google apps instead, see our guide to the Claude for Google Docs, Sheets and Slides add-on.
Frequently Asked Questions
How much does Claude Haiku 5.5 cost?
Up to 100K tokens, it’s $0.10 per million input tokens and $0.50 per million output tokens. Over 100K tokens, those rates rise to $0.50 and $2.50.
Is Claude Haiku 5.5 cheaper than GPT-6 Luna?
They match at the base rates. VentureBeat notes Luna’s higher rates start above 272K input tokens, while Haiku’s start at 100K, so your prompt size decides the winner.
What is the Claude Haiku 5.5 API model ID?
It’s claude-haiku-5-5. Anthropic says it’s live on its own platform plus AWS, Google Cloud and Azure.
Can I use Haiku 5.5 in the Claude app?
Anthropic’s launch page covers API platforms and doesn’t mention the Claude app. Check your model picker, since availability may differ by plan.
Our take
Haiku 5.5 is a strong pick for high-volume work, but read the 100K rule before you migrate anything. Run a week of real traffic through it, log your prompt sizes, and compare the bill against your old Haiku 4.5 spend. If most prompts stay short, you’ll likely save a lot. If they run long, Sonnet or Luna may be the smarter spend.
Leave a Reply