Claude Haiku 4.5
Kept on the board so the Claude ladder is honest. Index 17. Useful only when the call is easy and the surrounding stack is already Claude.
17Index · Published
$0.28Per measured task
60.7 pts/$Points / $
91Tokens / sec
No High score is published. Showing Published.
What this workload costs
List price for one call. Reasoning tokens count as output. A cache hit uses the lab’s cache rate when one is published.
Claude Haiku 4.5
Cache appliedSpecs
- API id
- claude-haiku-4-5
- Released
- Oct 2025
- Context
- 200k
- Max output
- 64k
- Input / 1M
- $1.00
- Output / 1M
- $5.00
- Cache read / 1M
- $0.10
- Knowledge
- Feb 2025
- Inputs
- Text, Image
- Weights
- Closed
The small Claude. GPT-6 Luna and the open flash models undercut it on both price and, for several of them, the index.
Use it when
A Claude-shaped API for routing and short calls, if you already live in Anthropic and the score of 17 is enough.
Leave it when
Anything you would hand to Sonnet, Sol, or an open flash model. Haiku is the previous generation's small model.