Quick facts
- Founded
- 2021
- Founders
- Dario Amodei, Daniela Amodei
- Headquarters
- San Francisco, California
- Flagship Model
- Claude 3.5 Sonnet
- Core Methodology
- Constitutional AI
Overview
Anthropic was established in 2021 as a Public Benefit Corporation (PBC) by former OpenAI research executives Dario and Daniela Amodei, along with key authors of GPT-3 and scaling laws research. The founders prioritized structural alignment research to ensure large foundation models remain steerable, honest, and harmless.
In 2022, Anthropic invented Constitutional AI, a training method where models critique and refine their own outputs against an explicit set of behavioral principles (a constitution) rather than relying exclusively on extensive manual human feedback. In 2024, Claude 3.5 Sonnet achieved industry-leading benchmarks in coding, reasoning, and multi-modal vision comprehension.
Company Milestones
- 2021Anthropic Founded
Established in San Francisco as an AI safety public benefit corporation.
- 2022Constitutional AI Paper
Published foundational methodology for training harmless AI assistants via self-critique.
- 2023Claude 1 & Claude 2
Introduced 100,000-token context windows and general availability on web and API.
- 2024Claude 3 & 3.5 Sonnet
Released Claude 3.5 Sonnet, establishing state-of-the-art benchmarks in software engineering and reasoning.
Forest Trails
- Founded By
- Flagship AI Models
- Methodology
- Origin Context
