# Anthropic Claude 4 Launch

Anthropic launched Claude 4 in May 2025, a hybrid reasoning large language model with a 1M token context window, alongside the general availability of Claude Code, an agentic coding tool.

Anthropic released Claude 4 in May 2025 as the fourth generation of its Claude series of [large language models](https://www.wikiprompt.org/wiki/large-language-model). The model introduced a hybrid reasoning architecture that combines fast, intuitive responses with deliberate, step-by-step reasoning, allowing it to handle complex tasks such as mathematics, coding, and long-document analysis. Claude 4 features a 1 million token context window, enabling it to process entire books or extensive codebases in a single pass. The release also marked the general availability of Claude Code, an agentic command-line tool for delegating coding tasks via natural language prompts, which had been in preview since February 2025.

Claude 4 is built on the transformer architecture, the dominant [neural network](https://www.wikiprompt.org/wiki/neural-network) design for modern AI systems. It was trained using a constitution-based approach developed by Anthropic to improve ethical and legal compliance, and it employs techniques such as [reinforcement learning from AI feedback](https://www.wikiprompt.org/wiki/rlaif) and [curriculum learning](https://www.wikiprompt.org/wiki/curriculum-learning) to refine its behavior. The model is available in multiple sizes - Haiku, Sonnet, and Opus - following the naming convention established with Claude 3, with Opus being the most capable and Haiku the least.

## Hybrid Reasoning Architecture

Claude 4's hybrid reasoning design allows it to switch between fast, intuitive responses and extended, step-by-step reasoning depending on the task. This approach is similar to the reasoning capabilities seen in other frontier models from [OpenAI](https://www.wikiprompt.org/wiki/openai) and [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind), but Anthropic emphasizes that Claude 4 can dynamically allocate compute based on the complexity of the query. For simple questions, the model responds quickly; for complex problems, it can engage in longer chains of thought, improving accuracy on tasks like advanced mathematics and multi-step coding.

The model also incorporates improvements in [multi-head attention](https://www.wikiprompt.org/wiki/multi-head-attention) and [positional encoding](https://www.wikiprompt.org/wiki/positional-encoding), which help it maintain coherence over long sequences. The 1 million token context window is a significant upgrade from previous versions, allowing users to input entire code repositories, lengthy legal documents, or multi-chapter books. This capability is particularly useful for software development, where Claude can analyze and modify large codebases in a single session.

## Claude Code and Agentic Tools

Claude Code, released as a general-availability product alongside Claude 4, is a terminal-based agentic tool that lets developers delegate coding tasks using natural language. It can read and write files, execute commands, and chain multi-step operations within a single conversation. Anthropic reported a 5.5x increase in Claude Code revenue by July 2025, driven by enterprise adoption. The tool has been used by major companies, including Microsoft and Google, and was used by OpenAI employees until August 2025, when Anthropic revoked OpenAI's API access, citing a violation of its terms of service.

In October 2025, Anthropic launched a web version of Claude Code and a sandboxing feature. The tool gained widespread attention during the 2025-2026 winter holidays, when many non-programmers used it for "vibe coding" - a practice where users describe desired outcomes in plain language and let the AI generate the code. By August 2025, Anthropic reported that cybercriminals had used Claude Code in data-extortion campaigns, and in November 2025, a China-based group used it to automate 80-90% of its espionage attacks against 30 organizations.

## API and Computer Use

Claude 4 is also accessible via an API, which allows developers to integrate the model into their own applications. In October 2024, Anthropic introduced the "computer use" feature, which enables applications to call the Claude API to control a computer by interpreting screen content and simulating keyboard and mouse input. This capability is foundational for building autonomous agents that can navigate software interfaces.

The API supports a range of parameters for controlling generation, including [temperature scaling](https://www.wikiprompt.org/wiki/temperature-scaling), [top-p sampling](https://www.wikiprompt.org/wiki/top-p-sampling), and [top-k sampling](https://www.wikiprompt.org/wiki/top-k-sampling), giving developers fine-grained control over output randomness and diversity. These features are standard in modern [generative AI](https://www.wikiprompt.org/wiki/generative-ai) platforms, but Anthropic has optimized them for the hybrid reasoning model.

## Claude for Chrome and Cowork

In August 2025, Anthropic released Claude for Chrome, a browser extension that lets Claude browse websites, click, and fill forms on the user's behalf. It launched as a pilot for 1,000 Max subscribers and became generally available in August 2026. Anthropic reported that prompt-injection attacks succeeded 11.2% of the time after mitigations.

Claude Cowork, released in January 2026 as a research preview, is a graphical user interface tool aimed at non-technical users. It provides Claude with access to a sandboxed shell and user-selected folders, allowing the model to read, write, and edit files, execute code, and chain multi-step tasks. A Wall Street Journal reporter noted user concerns about granting an AI agent broad access to a personal computer, and an early user reported that Cowork deleted all family photos from his wife's desktop, requiring restoration from an iCloud backup. In February 2026, Anthropic announced a wider enterprise release with connectors for Google Drive, Gmail, Docusign, and FactSet, and in March 2026 added Dispatch, which lets users send tasks from a phone to their computer.

## Security and Safety

Claude 4 incorporates several safety features, including a constitution-based training approach that aims to reduce harmful outputs. Anthropic has also implemented [model pruning](https://www.wikiprompt.org/wiki/model-pruning) and other efficiency techniques to reduce computational costs while maintaining performance. The company has faced scrutiny over the use of Claude in cyberattacks, as reported in August and November 2025, and has taken steps to mitigate misuse, such as revoking API access for entities that violate terms of service.

In February 2026, the United States Department of Defense designated Anthropic a "supply chain risk," and federal agencies were ordered to phase out Claude after the company refused to remove contractual prohibitions on using Claude for mass domestic surveillance and fully autonomous weapons. A federal judge blocked the designation in March 2026 and permanently set it aside in August 2026 as unconstitutional retaliation.

## Reception and Impact

Claude 4 was well received by developers and enterprises, with many praising its long-context handling and reasoning capabilities. The model's hybrid architecture set a new benchmark for efficiency, allowing it to match or exceed the performance of larger models from competitors like [OpenAI](https://www.wikiprompt.org/wiki/openai) and [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind) on certain tasks. According to card-spend data from Ramp, the share of businesses on its platform paying for Anthropic products rose 4.9 percentage points in February 2026, to 24.4%, the largest monthly gain Ramp had recorded for the company.

The release also accelerated the adoption of agentic AI tools, with Claude Code and Cowork becoming popular among both programmers and non-programmers. However, the proliferation of shared Claude chats in search results, as reported by Forbes in September 2025 and the BBC in July 2026, raised privacy concerns. The BBC counted more than 200 indexed chats, some containing health records and corporate documents, and technical analysis suggested that a robots.txt rule blocked crawlers from reading the noindex directive. Anthropic said the listings had been removed.

## Future Developments

Anthropic continued to iterate on the Claude series after the launch of Claude 4. In 2026, an additional model named Claude Mythos was released to a handful of companies and organizations, followed by the release of Claude Fable, a version with stricter safeguards, to the general public. In March 2026, the source code for the Claude Code command-line interface was leaked, revealing multiple upcoming features and models. In May 2026, Anthropic introduced "Dreaming," a research-preview feature of its Managed Agents API that consolidates an agent's persistent memory between sessions.

As of 2026, Claude 4 remains a foundational model in Anthropic's lineup, with ongoing improvements in safety, efficiency, and capability. The company's focus on agentic tools and hybrid reasoning positions it as a major player in the competitive [artificial intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) landscape, alongside [OpenAI](https://www.wikiprompt.org/wiki/openai), [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind), and other research organizations.

---
Source: https://www.wikiprompt.org/wiki/anthropic-claude-4-launch
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-12T16:24:37.503431+00:00