- Haiku 5.5 launched on 7 October 2026 at $0.10 input and $0.50 output per million tokens for prompts up to 100K tokens.
- Haiku 4.5 was $1 and $5. Anthropic says the new model costs about 75% less on average, and 90% less for requests up to 100K tokens.
- Past 100K tokens in one prompt, the price is $0.50 in and $2.50 out. That is still about half of Haiku 4.5, but 5x the lower tier.
- Token counts can differ between models, so test your own prompts before you plan a budget.
The new prices
Anthropic released Claude Haiku 5.5 on 7 October 2026. Here is how the list prices compare, per million tokens.
| Model | Prompt size | Input | Output |
|---|---|---|---|
| Haiku 4.5 | Any size | $1.00 | $5.00 |
| Haiku 5.5 | Up to 100K tokens | $0.10 | $0.50 |
| Haiku 5.5 | Over 100K tokens | $0.50 | $2.50 |
75% or 90%? Both are right
You will see both numbers online, and they look like they disagree. They do not. Anthropic says Haiku 5.5 costs around 75% less to run on average. A footnote on the same page says it is priced 90% lower than Haiku 4.5 for requests up to 100,000 tokens, and 50% lower above that.
So the 90% figure is for shorter prompts, and the 75% figure is the average across typical use. Which one applies to you depends on how long your prompts are.
The catch: the 100K line
The low price applies to prompts up to 100K tokens. Go past that in a single prompt and the rate rises to $0.50 in and $2.50 out. That is five times the lower tier of Haiku 5.5 itself, though still about half of what Haiku 4.5 cost.
Will the same task use more tokens?
Possibly a little. Anthropic notes that Haiku 5.5 uses an updated tokenizer, so the same text can turn into slightly more tokens, and says its savings figures take that into account. One independent test of a long prompt reported about 1.25x more tokens than on Haiku 4.5.
That is one test on one prompt, so treat it as a warning, not a rule. The saving is still large. It is just wise to measure your own prompts instead of trusting a headline.
Sonnet 5.5 cache reads got cheaper too
Prompt caching lets you reuse text you send often, such as a long instruction block or a product catalogue. For Sonnet 5.5, cache reads now cost $0.10 per million tokens, down from $0.20. If your workflow sends the same large context again and again, this matters as much as the headline price.
Try the maths on your own volume
Here is a simple example. Say 1,000 customer replies, each with 1,500 tokens in and 300 tokens out. At list prices that is about $3.00 on Haiku 4.5 and about $0.30 on Haiku 5.5. Even if Haiku 5.5 used 25% more tokens for the same work, it would be about $0.38. Change the numbers below to match your own.
Cost calculator
Illustrative only. Uses list prices for prompts up to 100K tokens.
Haiku 4.5: $1 in, $5 out per million tokens. Haiku 5.5: $0.10 in, $0.50 out. The multiplier is applied to both input and output tokens as a simple assumption. Your real bill will vary.
What it means for a business like yours
When AI work gets this cheap, jobs that were not worth automating start to make sense. Think of the high-volume, low-risk tasks: sorting incoming enquiries, drafting a first reply, summarising a long chat before your team calls back, or tagging leads by what they asked for.
A cheaper model does not replace good judgement. Anything a customer will read should still be checked by a person, especially in sensitive areas like health. But the cost of trying has dropped a lot, and that is the real news.
If AI got ten times cheaper tomorrow, which task in your business would you hand over first? Tell us, and we will give you an honest view on whether it is worth automating, including when it is not.
Sources
- Anthropic: Claude Haiku 5.5 launch page (prices, the 75% and 90% figures, tokenizer note, Sonnet 5.5 cache reads)
- Anthropic: Claude Haiku product page (pricing tiers)
- Independent test of token counts: as reported by one developer on a single long prompt. Run your own comparison before you decide.
The worked example is simple arithmetic on list prices, not a quote or a guarantee. Prices can change, so check Anthropic's pricing page before you budget.