Claude Haiku 5.5 cuts its own price and closes the gap on bigger models

Anthropic has released Claude Haiku 5.5, which it calls its cheapest, fastest, and most capable small model so far. The headline is the price: input drops to $0.10 per million tokens from $1.00 on Haiku 4.5, output to $0.50 from $5.00, which Anthropic says works out to about 75% cheaper on average. A new tokenizer uses slightly more tokens per task, and that cost is already folded into the figure.

The benchmark jumps are what make the price interesting. On Terminal-Bench 4.0 the old Haiku 4.5 scored 0.0%; Haiku 5.5 scores 39.2%. On the OSWorld computer-use test it goes from 15.7% to 72.4%. These are large gains for a model that costs a fraction of the previous one. Haiku 5.5 is also the first model in the Haiku line with an adjustable effort setting, from Low to Max, so you can trade spend for intelligence on a given task instead of switching models.

It is still a small model, and Anthropic is clear about that. Sonnet 5.5 reaches 70.6% on Terminal-Bench 4.0 against Haiku's 39.2%, and remains the recommendation for complex agentic coding. Alongside the launch, Anthropic halved Sonnet 5.5's cache-read price, which it says cuts the cost of most agentic work with that model by about 20%.

Why it matters

If you run high-volume, cost-sensitive work like classification, summaries, or subagent calls, the math for pushing more of it onto a small model just changed. Test Haiku 5.5 at a higher effort setting against whatever you currently use for those jobs before assuming you still need a mid-tier model.