Claude 3 (Opus, Sonnet, Haiku)
Claude 3 launches as a three-tier family (Opus, Sonnet, Haiku) with image input, 200K context and fewer refusals; Opus and Sonnet ship first.
First Claude models that read images, charts and diagrams. 200K context standard, 1M-token inputs offered to select customers. Opus claims >99% needle-in-a-haystack recall. Tiered pricing per million tokens is Opus $15/$75, Sonnet $3/$15 and Haiku $0.25/$1.25. Models refuse far fewer borderline prompts.
- Date
- Monday, 4 March 2024
- Lab
- Anthropic
- Kind
- model
- Access
- closed API
- Price
- Opus $15/$75, Sonnet $3/$15, Haiku $0.25/$1.25 per M tokens in/out (2024-03-04)
Figures
| Measure | Value | Measured by |
|---|---|---|
| Needle-in-a-haystack recall (Opus) | >99% 200K context | company |
| Context window | 200K tokens 1M-token inputs for select customers | company |
MMLU, GPQA, GSM8K and HumanEval comparisons are shown in charts on the page; the values were not available as text, so none are listed. Haiku became available 2024-03-13.
Sources
This record was checked against its sources on 6 October 2026. How we check