Anthropic introduces the Claude 3 model family

Background

Claude 3 was introduced as a family of three models rather than a single model. Each tier was aimed at a different balance of speed, cost and capability.^[1]

What happened

On 4 March 2024 Anthropic announced the Claude 3 family. It has three models, Claude 3 Haiku, Claude 3 Sonnet and Claude 3 Opus, which rise in capability in that order.^[1]

Opus and Sonnet launched in claude.ai and the API in 159 countries. Sonnet powered the free tier of claude.ai, while Opus was offered to Pro subscribers. Haiku followed later.^[1]

Key details

Prices per million input and output tokens were $15 and $75 for Opus, $3 and $15 for Sonnet, and $0.25 and $1.25 for Haiku. A token is a small piece of text, roughly a word or part of a word.^[1]

All three models launched with a 200K-token context window. Anthropic said they could accept inputs over 1 million tokens, but this was offered to select customers.^[1]

The models added vision. They can handle photos, charts, graphs and technical diagrams.^[1]

Sonnet was available on Amazon Bedrock at launch and in private preview on Google Cloud Vertex AI.^[1]

Anthropic said the models remained at AI Safety Level 2 (ASL-2) under its Responsible Scaling Policy.^[1]

Impact

The family gave users a choice of price and capability within one line of models, and all three tiers added vision.^[1]

Vision here means the model can take images as input and reason about them. It does not mean the model generates pictures. Pricing per million tokens means the cost of using a model scales with how much text a user sends and receives.

The Responsible Scaling Policy is Anthropic's framework for tying safety measures to the capability of its models. AI Safety Level 2 is one of the levels in that framework. Keeping all three models at that level meant Anthropic judged the family to need the same tier of safeguards.^[1]