Anthropic’s Claude 3 family adds vision and a faster range of models

Anthropic announced three Claude 3 models today: the compact Haiku, balanced Sonnet and most capable Opus. Sonnet is available immediately through Claude and the company’s programming interface; Opus is also available, with Haiku promised soon.

All three accept images as well as text and support a 200,000-token context window. Anthropic reports improved results across reasoning, mathematics and coding evaluations, and says the models are less likely than earlier Claude versions to refuse harmless questions.

A model family lets organizations choose speed and price for each task rather than using one system everywhere. Vision support also opens document, chart and photograph analysis, though Anthropic limits some processing of people and warns that outputs can still be inaccurate. Independent testing will show how benchmark advantages translate into daily work and whether the largest model’s extra cost pays off.