Research note
Research note: Anthropic releases Claude Haiku 5.5, cutting small-model API prices by up to 90%
How we checked Claude Haiku 5.5’s prices, availability and benchmark claims against Anthropic’s page and VentureBeat.
What we set out to check
The price per token, where the model is available, what Anthropic claims on benchmarks and safety, and what is missing.
What we found
- Anthropic’s page and VentureBeat agree on all prices.
- Benchmarks come only from Anthropic.
Numbers that differ
The headline “90% cheaper” applies to prompts up to 100,000 tokens. Above that the cut is 50%, and Anthropic itself estimates about 75% lower average cost because of a new tokenizer. We say “up to 90%” in the headline and explain the rest.
Labels we chose and why
- Confirmed for prices and availability.
- Company claim for benchmark scores and the 75% estimate.
- We left out a reported prompt-injection claim we could not find on Anthropic’s page.
What we could not verify
- The context window, which the launch page does not state.
- Independent benchmark results.
Disclosure
Anawords uses Claude as a research tool. We applied the same rules to this story as to any other.
What we searched
- Claude Haiku 5.5 launch Anthropic
- Claude Haiku 5.5 price per million tokens
- Claude Haiku 5.5 benchmarks OSWorld
Every source we read
| Source | Used? | Why |
|---|---|---|
| Anthropic: Claude Haiku 5.5 | USED | Primary source for prices, availability, benchmarks and safety settings. |
| VentureBeat | USED | Independent report. Confirms prices and adds the comparison with GPT-6 Luna pricing. |
| Help Net Security | NOT USED | Headline refers to prompt-injection gains that we could not confirm on Anthropic’s page. |