Anthropic
Anthropic was founded in 2021 by Dario Amodei, Daniela Amodei, and seven other former OpenAI employees who left over disagreements about safety practices and governance. It is structured as a public benefit corporation and has built its brand around Constitutional AI — a technique for training models to be helpful, harmless, and honest using a set of principles rather than solely human feedback. Anthropic's Claude model family is widely regarded as a serious technical rival to GPT-4. The company has raised over $7 billion from Google and Amazon. It occupies an unusual position in the industry: genuinely concerned about frontier AI risk while simultaneously accelerating frontier AI development. AINSI tracks Anthropic's Responsible Scaling Policy as the most detailed public safety commitment from any frontier AI lab.
AI Models
Claude 3.5 Sonnet
Released in June 2024, Claude 3.5 Sonnet achieved the highest scores on coding benchmarks among commercial models at launch and became the primary model powering Anthropic's Claude.ai. It introduced Computer Use — the ability to operate a desktop computer by taking screenshots and issuing mouse and keyboard commands — in October 2024. Claude 3.5 Sonnet is the model AINSI uses for research assistance and document analysis. Anthropic's Constitutional AI training methodology is most visible in this model's refusal patterns and reasoning transparency.
Claude 3 Opus
Released in March 2024, Claude 3 Opus was Anthropic's highest-capability model at launch — outperforming GPT-4 on most professional benchmarks. It was succeeded by Claude 3.5 Sonnet six months later. Opus remains notable for its performance on complex analytical tasks requiring sustained multi-step reasoning and its unusually high scores on the MMLU graduate-level knowledge benchmark.