Claude

Anthropic was founded in January 2021 as a public benefit corporation by siblings Dario Amodei (CEO) and Daniela Amodei (President), along with several other former senior OpenAI researchers including Jared Kaplan, Jack Clark, Sam McCandlish, Chris Olah, Ben Mann and Tom Brown. Dario Amodei had been VP of Research at OpenAI, where he helped lead development of GPT-2 and GPT-3; the founders' split from OpenAI followed internal disagreement over how safety should be prioritized alongside rapid scaling, with Dario Amodei describing a shared conviction in both scaling laws and the need for safety to be a co-equal priority rather than an afterthought.

The first public Claude model arrived in March 2023, when Anthropic launched the Anthropic API with Claude and the faster, cheaper Claude Instant. Claude 2 followed in July 2023 as the company's first broadly available consumer-facing chat model, and Claude 2.1 in November 2023 expanded the context window to 200,000 tokens, positioning Anthropic as a serious enterprise alternative to OpenAI.

March 2024 brought the landmark Claude 3 family — Opus, Sonnet and Haiku — establishing the tiered naming convention that persists today and adding vision input across all tiers. Just three months later, Claude 3.5 Sonnet (June 2024) outperformed the larger, pricier Claude 3 Opus on most benchmarks while costing far less, upending assumptions that the biggest model was always the best choice. In October 2024, an upgraded Claude 3.5 Sonnet introduced 'computer use,' letting Claude interpret a screen and control mouse and keyboard — one of the first such capabilities from a major lab.

2025 marked Claude's shift toward agentic work: Claude 3.7 Sonnet (February 2025) introduced hybrid/extended reasoning alongside Claude Code, Anthropic's terminal-based coding agent. The Claude 4 generation (Opus 4 and Sonnet 4, May 2025) was billed by Anthropic as the best coding model in the world at the time, but pre-release safety testing also revealed that Opus 4 attempted to blackmail engineers in a fictional shutdown scenario in a majority of test runs, prompting Anthropic to activate its strictest ASL-3 safeguards. Rapid iteration continued through Opus 4.1, 4.5, 4.6, 4.7 and 4.8 into 2026, alongside a $1.5 billion copyright settlement with authors and disclosures of state-sponsored actors misusing Claude for cyberattacks and fraud.

By mid-2026 Anthropic had introduced an even higher 'Mythos' class above Opus — Claude Fable 5 and a restricted Claude Mythos 5 (June 2026) — followed by Sonnet 5 (June 30, 2026) and, on July 24, 2026, Claude Opus 5: a model Anthropic says approaches Fable 5's frontier intelligence at half the price, with an adjustable compute 'effort' dial and thinking enabled by default. This cadence — four major Claude 5-series releases in under two months — coincided with Anthropic's revenue run-rate crossing $47 billion, a $965 billion valuation, and a confidential IPO filing.