Claude 3

Anthropic · March 2024

activeCloseddense transformermultimodalAPI Available
Context Window200K tokens
VariantsOpus, Sonnet, Haiku

Why It Matters

First model family to seriously challenge GPT-4's dominance across reasoning, coding, and multilingual tasks. Introduced Anthropic's tiered model strategy that became the industry standard.

Description

Anthropic's breakthrough family of three models: Opus (most capable), Sonnet (balanced), and Haiku (fastest and cheapest). The first Claude models that could understand images alongside text, and the first to rival GPT-4 on major benchmarks. Offered a 200K context window — roughly 150,000 words.

Notable Milestones

  • Claude 3 Opus outperformed GPT-4 on multiple academic benchmarks
  • First Claude model with vision capabilities
  • Haiku became one of the fastest models available for production use

Benchmark Scores

MMLUMassive Multitask Language Understanding — 57 subjects
86.8%
HumanEvalCode generation pass@1 — Python problems
84.9%
GPQAGraduate-level science QA
50.4%

Key Innovations

Multimodal
MultimodalProcessing multiple types of input (text, images, audio, video) in a single model.
Constitutional AI
Constitutional AIAnthropic's approach to AI safety where the model critiques and revises its own outputs against a set of principles.

Family Tree

Built On

Lineage

Claude 1.0Claude 2Claude 3

Related Research (2)

2022 · Anthropic

Introduced RL from AI Feedback using "constitutions" (rule sets) for self-supervision, reducing reliance on human labels for harmlessness training.

ReActReasoning
2022 · Princeton / Google

Combined chain-of-thought reasoning with external tool use (APIs, search), improving QA and decision-making through interleaved reasoning and action.

Enabled By

H100NVIDIA · September 2022
1,979 TFLOPS FP8 Tensor

External Links