Claude 3 Haiku
Claude 3 Haiku, the fastest and cheapest Claude 3 model, reaches the API and Claude Pro, reading about 21K tokens per second on short prompts.
Aimed at high-volume enterprise workloads. Anthropic claims about 3x the throughput of peers in its tier for most workloads, 21K tokens (about 30 pages) per second for prompts under 32K tokens. Priced $0.25/$1.25 per million tokens and keeps vision input.
- Date
- Wednesday, 13 March 2024
- Lab
- Anthropic
- Kind
- model
- Access
- closed API
- Price
- $0.25/$1.25 per M tokens in/out (2024-03-04 family price)
Figures
| Measure | Value | Measured by |
|---|---|---|
| Throughput on prompts under 32K tokens | 21K tokens/s 30-60% slower above 32K tokens | company |
Vertex AI availability was 'coming soon' on the launch page.
Sources
This record was checked against its sources on 6 October 2026. How we check