Anthropic Claude Haiku 4.5
Anthropic has released Claude Haiku 4.5 (claude-haiku-4-5-20251001) their latest small model which is good for low-latency tasks where you need speed/cost more than being accurate.
In terms of benchmark it is as good as Claude Sonnet 4.
Based on the benchmark, it looks like good for agentic use case where this model is used as sub-agent in combination with an orchestrator model like Sonnet 4.5.
And here is the comparison between other models
The pricing is more than old Haiku (but I guess they increased it because of intelligence?)
| Token Type | Price per million |
|---|---|
| Input | $1 |
| Output | $5 |
Sam Witteveen has a high level walk through of this model in his recent video 👇
Reference
