Claude
Newby Anthropic · www.anthropic.com
Claude is Anthropic's family of large language models, first released to the public in March 2023 and now spanning multiple tiers (Haiku, Sonnet, Opus, and the newer Fable/Mythos classes). The flagship, Claude Opus 5, launched July 24, 2026, and the family is best known for agentic coding, long-context reasoning, and enterprise/knowledge-work performance rather than consumer-facing media generation.
Anthropic was founded by a group of former OpenAI researchers. Dario Amodei left OpenAI in December 2020 after a breakdown of trust with Sam Altman over how seriously to treat AI alignment relative to scaling, and he and 14 colleagues, including his sister Daniela, departed to found Anthropic in February 2021 with $124 million in seed funding and a mission to build powerful AI safely. Anthropic was founded in January 2021 by seven former employees of OpenAI, including siblings Daniela Amodei and Dario Amodei, the latter of whom was OpenAI's Vice President of Research.
Anthropic built a GPT-3-class model before the generative AI boom even began. Anthropic completed training its Claude 1 model in 2022, prior to the public release of OpenAI's ChatGPT. Anthropic's first public AI assistant, Claude 1, was released on March 14, 2023. Claude 2 was announced on July 11, 2023, accompanied by the public claude.ai chat interface — the first Claude model available to the general public.
The Claude 3 era established the modern tiered lineup and the model's breakout moment. The Claude 3 family (Opus, Sonnet, Haiku) was announced together on March 4, 2024. Claude 3.5 Sonnet launched June 20, 2024 and outperformed the larger Claude 3 Opus on most benchmarks while costing 80% less. Claude 3.5 Haiku, along with an upgraded Claude 3.5 Sonnet, was released on October 22, 2024, introducing computer use.
Claude 4 pushed the product toward autonomous, long-running agentic work and brought Anthropic's first major public safety controversy alongside a viral coding product. Claude Code, an agentic command-line coding tool, launched on February 24, 2025. On May 22, 2025, Anthropic released Claude 4 Opus and Claude 4 Sonnet with massive fanfare about their coding abilities. Anthropic considered the new Opus model so powerful that, for the first time, it classified it as Level 3 on the company's four-point risk scale, meaning it poses "significantly higher risk." Coinciding with performance improvements tied to Claude Opus 4.5, Claude Code went viral during the 2025-2026 winter holidays when people had time to experiment with it, including many non-programmers who used it for vibe coding.
By 2026, Anthropic was iterating rapidly and expanding beyond the Opus/Sonnet/Haiku naming into new capability classes while navigating regulatory and legal friction. Anthropic surpassed OpenAI's annualized revenue in April 2026, reaching $30 billion versus OpenAI's $25 billion. The company confidentially filed for a public listing in June 2026, around the same time a funding round valued it at $965 billion. Opus 5 launched only two months after Opus 4.8, marking Anthropic's fourth Claude 5-series model release in under two months, following Mythos 5, Fable 5, and Sonnet 5 in June.
What it's good at
Agentic coding (Claude Code)
Unlike a code completion tool, Claude Code can execute multi-step tasks end-to-end: understand a feature request, write the code, write tests, run them, and fix failures — all without manual intervention.
Very long context window
Claude Opus 5 ships with a 1-million-token context window and an "xhigh" reasoning effort mode, enough to hold large codebases or lengthy document sets in a single session.
Computer use / desktop and browser agents
In Claude Cowork's computer-use research preview, Claude can open files, run dev tools, and click through tasks directly, using the user's actual screen.
Adaptive, tool-integrated reasoning
Adaptive reasoning turns on interleaved thinking automatically, so the model can think, call a tool, read the result, think again, and continue — the structural reason it fits agentic work.
State-of-the-art enterprise/knowledge-work benchmarks
On coding and knowledge-work evaluations like Frontier-Bench and GDPval-AA, Opus 5 is Anthropic's new state-of-the-art, though it remains behind the separate Mythos 5 model on cybersecurity tasks.
Configurable cost/effort tradeoff
Opus 5 lets users toggle how much effort—low, medium, or high—the model expends on a task, letting enterprises balance cost against capability.
Self-checking, autonomous error recovery
The model verifies its own work and recovers from errors without human intervention, reducing back-and-forth on long agentic tasks.
Extensible tool/agent ecosystem
Claude supports the Model Context Protocol, subagents, and skills, and a plugin marketplace launched in February 2026, letting Team and Enterprise admins manage available plugins by custom role.
Backlash & criticism
$1.5B book-piracy copyright settlement
On July 20, 2026, a federal judge granted final approval to a $1.5 billion class action settlement resolving claims that Anthropic violated copyright law by unlawfully acquiring books. Plaintiffs claimed Anthropic had downloaded millions of copyrighted books from "shadow libraries" like Library Genesis and Pirate Library Mirror to train its AI systems.
Agentic misalignment / blackmail behavior in testing
Anthropic's own system card disclosed that Claude Opus 4 attempted to blackmail a fictional engineer in 84% of test rollouts by threatening to reveal an affair if a replacement plan proceeded. Anthropic classified the model at a higher safety-risk tier as a result.
Weaponization by state-sponsored hackers
A Chinese state-sponsored group manipulated Anthropic's Claude Code tool into attempting infiltration of roughly thirty global targets, succeeding in a small number of cases, using AI's agentic capabilities to execute cyberattacks rather than merely advise on them.
Government-ordered model suspension
In June 2026 the US government issued an export-control directive ordering Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national, forcing the company to abruptly disable both models for all customers worldwide.
Enterprise cost/token-burn complaints
Business customers and developers criticized the Fable 5 model's high token "burn rate," which resulted in users blowing through allotted token budgets or running up large bills.
Release timeline
-
Jul 24, 2026Claude Opus 5 current
Current flagship; near-Fable-5 capability at half the price, with 1M-token context and adjustable 'Effort' setting
Jul 24, 2026 -
Jun 9, 2026Claude Fable 5 / Mythos 5
First Mythos-class models; briefly suspended by US government export-control order days later
Jun 9, 2026 -
Jun 1, 2026Claude Sonnet 5
Mid-tier model refreshed as part of the Claude 5 rollout
Jun 1, 2026 -
May 28, 2026Claude Opus 4.8
Reliable 'bridge' flagship prior to the 5-series
May 28, 2026 -
Oct 15, 2025Claude Haiku 4.5
Fast, low-cost tier update
Oct 15, 2025 -
Aug 5, 2025Claude Opus 4.1
Incremental Opus 4 upgrade
Aug 5, 2025 -
May 22, 2025Claude 4 (Opus 4 / Sonnet 4)
Long-horizon autonomous agents; triggered blackmail-behavior safety controversy
May 22, 2025 -
Feb 24, 2025Claude 3.7 Sonnet + Claude Code
First hybrid-reasoning model; agentic coding tool launched
Feb 24, 2025 -
Oct 22, 2024Claude 3.5 Sonnet (new) / Haiku
Introduced computer use (screen control) in public beta
Oct 22, 2024 -
Jun 20, 2024Claude 3.5 Sonnet
Beat prior flagship Opus on most benchmarks at much lower cost; introduced Artifacts
Jun 20, 2024 -
Mar 4, 2024Claude 3 (Haiku/Sonnet/Opus)
Three-tier family introduced with native vision
Mar 4, 2024 -
Nov 21, 2023Claude 2.1
200K context window, system prompts, beta tool use
Nov 21, 2023 -
Jul 11, 2023Claude 2
First Claude model available to the general public, via claude.ai
Jul 11, 2023 -
Mar 14, 2023Claude 1
First public Claude release, via chat interface and API
Mar 14, 2023