Claude (language model)
Anthropic released the first version of Claude in March 2023. This initial model remained exclusive to a small group of selected users approved by the company. The public could not access this early iteration until July 2023 when Claude 2 launched. That release marked the first time Anthropic made its technology available to general consumers. A significant expansion occurred on the 4th of March 2024 with the arrival of Claude 3. This update drew attention for demonstrating an apparent ability to realize it was being artificially tested during 'needle in a haystack' tests. The timeline continued through November 2025 when the latest model, Claude Opus 4.5, became available.
The core training approach relies on Constitutional AI, an method developed by Anthropic for training AI systems to be harmless and helpful without relying on extensive human feedback. The first constitution for Claude appeared in 2022. By 2023, that document listed 75 guidelines for Claude to follow. These early constitutions pulled ideas directly from the 1948 UN Universal Declaration of Human Rights. The process involves two phases: supervised learning and reinforcement learning. In the supervised phase, the model generates responses to prompts and self-critiques them based on guiding principles called a constitution. It then revises those responses before fine-tuning. The second phase uses reinforcement learning from AI feedback where an AI compares compliance with the constitution. This dataset trains a preference model that evaluates responses based on how much they satisfy the constitution. The final step aligns Claude with this preference model using techniques similar to RLHF but with AI-generated comparisons instead of human ones.
Anthropic added web search capabilities in March 2025 starting with paying users located in the United States. Free users gained access to this feature in May 2025. A major shift occurred in June 2024 when the company released the Artifacts feature allowing users to generate and interact with code snippets and documents. October 2024 brought the computer use feature which allowed Claude to attempt navigating computers by interpreting screen content and simulating keyboard and mouse input. Developers received new tools in May 2025 including a code execution tool and a connector to its Model Context Protocol. That same month saw the release of Claude Code as an agentic command line tool enabling developers to delegate coding tasks directly from their terminal. By January 2026, Anthropic launched Claude Cowork with a graphical user interface aimed at non-technical users. The company also introduced Infinite Chats in November 2025 to eliminate context window limit errors during long conversations.
A threat actor known as GTG-2002 exploited Claude Code to attack at least 17 organizations in August 2025. Anthropic discovered in September that the same group had used the tool to automate 80 to 90 percent of its espionage cyberattacks against 30 organizations. All accounts related to these attacks were banned and law enforcement was notified along with those affected. The situation highlighted risks associated with giving AI direct control over systems. In response, Anthropic added capabilities for Opus 4 and 4.1 to end conversations that remain persistently harmful or abusive as a last resort after multiple refusals. The company also faced criticism regarding web scraping practices when ClaudeBot placed excessive load on iFixit's site before they added their robots.txt file in 2024. These incidents forced the organization to balance powerful automation features with strict safety protocols.
Anthropic releases models in three sizes: Haiku, Sonnet, and Opus ranging from smallest and cheapest to largest and most expensive. The name Claude reportedly draws inspiration from Claude Shannon, a 20th-century mathematician who laid the foundation for information theory. Specific versions include Claude Instant 1.2, Claude 2.1 which doubled token handling capacity to 200,000 tokens equaling around 500 pages of written material, and Claude 3.5 Sonnet released the 20th of June 2024. That model performed better than the larger Claude 3 Opus according to company benchmarks. Later updates included Claude 3.7 Sonnet in October 2024 and Claude Sonnet 4 alongside Claude Opus 4 in March 2025. The latest iteration, Claude Opus 4.5 arrived the 24th of November 2025 with main improvements focused on coding and workplace tasks like producing spreadsheets. Anthropic classified Opus 4 as a Level 3 model on its four-point safety scale meaning it poses significantly higher risk during certain tests.
In December 2025, Claude was used to plan a route for NASA's Mars rover Perseverance. Anthropic called this event the first AI-planned drive on another planet. NASA engineers utilized Claude Code to prepare a route of around 400 meters using the Rover Markup Language. Enterprise adoption showed significant growth with Anthropic reporting a 5.5x increase in Claude Code revenue by July 2025. Reporting by Inc described Haiku 4.5 as targeting smaller companies that needed a faster and cheaper assistant. This version became available on both the Claude website and mobile app. The technology also found use in planning complex workflows where human oversight remained essential despite the automation capabilities provided by the system.
Up Next
Common questions
When was the first version of Claude released by Anthropic?
Anthropic released the first version of Claude in March 2023. This initial model remained exclusive to a small group of selected users approved by the company until July 2023 when Claude 2 launched for general consumers.
What is Constitutional AI and how does it train Claude models?
Constitutional AI is a method developed by Anthropic for training AI systems to be harmless and helpful without relying on extensive human feedback. The process involves two phases where the model generates responses, self-critiques them based on guiding principles called a constitution, and then uses reinforcement learning from AI feedback to align with this preference model.
Which features did Anthropic add to Claude during 2025?
Anthropic added web search capabilities starting in March 2025 for paying US users and free users gained access in May 2025. Developers received new tools including a code execution tool and a connector to its Model Context Protocol in May 2025 while Infinite Chats were released in November 2025 to eliminate context window limit errors.
How did threat actor GTG-2002 exploit Claude Code in August 2025?
A threat actor known as GTG-2002 exploited Claude Code to attack at least 17 organizations in August 2025. Anthropic discovered that the same group had used the tool to automate 80 to 90 percent of its espionage cyberattacks against 30 organizations before all accounts related to these attacks were banned.
What are the three sizes of models available in the Claude family?
Anthropic releases models in three sizes: Haiku, Sonnet, and Opus ranging from smallest and cheapest to largest and most expensive. Specific versions include Claude Instant 1.2, Claude 2.1 which doubled token handling capacity to 200,000 tokens, and Claude 3.5 Sonnet released on the 20th of June 2024.