From the archive
Anthropic introduces Claude 3 with image understanding
Claude 3 launches with Opus and Sonnet, image inputs, a 200,000-token context window, and a planned Haiku release.
Anthropic announced the Claude 3 family on March 4, 2024. Claude 3 Opus and Sonnet became available through claude.ai and the Claude application programming interface (API), with Haiku scheduled to follow.
Three model sizes
The family separated larger, more capable models from a faster, lower-cost option. Opus was the top model in the initial range, Sonnet the middle option, and Haiku the smallest. That gave applications a choice based on the task and the cost of processing it.
Images and longer inputs
Claude 3 added image understanding for material such as photographs, charts, and technical diagrams. The initial models offered a 200,000-token context window, allowing longer documents and conversations to fit in a request.
Anthropic also reported changes intended to reduce unnecessary refusals and improve factual accuracy. The release expanded Claude beyond text-only interactions while retaining the need to check its answers.