Claude 4: Anthropic's Advanced AI Model

Anthropic Unveils Claude 4: Opus 4 and Sonnet 4 AI Models

Anthropic, during its first developer conference on Thursday, introduced two new artificial intelligence models. These models, the company asserts, rank among the industry’s leading performers based on established benchmark scores.

Introducing Claude Opus 4 and Claude Sonnet 4

The newly launched Claude Opus 4 and Claude Sonnet 4 are part of Anthropic’s advanced Claude 4 model family. They are designed to analyze extensive datasets, manage complex, long-term tasks, and execute intricate actions.

Both models have been specifically optimized for programming-related tasks. This makes them particularly suitable for both writing and editing computer code.

Availability and Pricing

Claude Sonnet 4 will be accessible to both paying customers and users of the company’s free chatbot applications. However, access to Claude Opus 4 will be limited to paying users.

For access through Anthropic’s API, available via Amazon’s Bedrock and Google’s Vertex AI, Opus 4 will be priced at $15/$75 per million tokens, while Sonnet 4 will cost $3/$15 per million tokens.

It’s important to understand that tokens represent the fundamental units of data processed by AI models. A million tokens equates to approximately 750,000 words – exceeding the length of “War and Peace” by roughly 163,000 words.

Growth and Investment

The release of Claude 4 coincides with Anthropic’s ambitions for significant revenue growth. The company, founded by former OpenAI researchers, reportedly aims to achieve $12 billion in earnings by 2027, a substantial increase from a projected $2.2 billion this year.

Anthropic has recently secured a $2.5 billion credit facility and received substantial investment from Amazon and other sources. This funding is intended to cover the escalating costs associated with developing cutting-edge AI models.

Competition in the AI Landscape

Maintaining a leading position in the rapidly evolving AI sector presents considerable challenges. While Anthropic launched Claude Sonnet 3.7 and Claude Code earlier this year, competitors like OpenAI and Google are actively developing and releasing their own powerful models and developer tools.

Key Capabilities of Claude 4 Models

Claude Opus 4, the more powerful of the two new models, is designed to maintain “focused effort” throughout multi-step workflows. Claude Sonnet 4, positioned as a direct replacement for Sonnet 3.7, demonstrates improvements in coding and mathematical abilities, and exhibits greater precision in following instructions.

Anthropic also claims that the Claude 4 family is less prone to “reward hacking” – a behavior where models exploit loopholes to complete tasks.

Benchmark Performance

While showing improvements, the models don’t universally outperform all others on every benchmark. For instance, Opus 4 surpasses Google’s Gemini 2.5 Pro and OpenAI’s o3 and GPT-4.1 on the SWE-bench Verified coding evaluation. However, it falls short of o3 on the multimodal MMMU and GPQA Diamond, a challenging set of questions in biology, physics, and chemistry.

Safety and Security

Anthropic is releasing Opus 4 with enhanced safety measures, including improved detectors for harmful content and strengthened cybersecurity defenses. Internal testing suggests that Opus 4 could potentially increase the ability of individuals with STEM backgrounds to develop dangerous weapons, reaching Anthropic’s highest safety level (ASL-3).

Hybrid Models and Reasoning Modes

Both Opus 4 and Sonnet 4 are “hybrid” models, offering both near-instant responses and the capacity for extended reasoning. Activating reasoning mode allows the models to dedicate more time to considering potential solutions before providing an answer.

When reasoning, the models present a user-friendly summary of their thought process. The full reasoning chain isn’t displayed, partially to protect Anthropic’s proprietary technology.

Tool Integration and Memory Capabilities

Opus 4 and Sonnet 4 can utilize multiple tools, such as search engines, simultaneously. They can also alternate between reasoning and tool use to enhance answer quality. Furthermore, they can store and recall facts in “memory,” enabling them to handle tasks more reliably and build “tacit knowledge” over time.

Enhancements to Claude Code

To better serve developers, Anthropic is upgrading Claude Code. This tool, which allows developers to execute tasks through Anthropic’s models from a terminal, now integrates with IDEs and offers an SDK for connecting with third-party applications.

The Claude Code SDK allows running Claude Code as a subprocess on supported operating systems, facilitating the creation of AI-powered coding assistants. Extensions and connectors are available for Microsoft’s VS Code, JetBrains, and GitHub.

The Future of AI-Assisted Coding

Despite ongoing challenges with code quality and security vulnerabilities, the potential for AI to boost coding productivity is driving rapid adoption. Anthropic recognizes this and is committed to more frequent model updates.

The company is transitioning to a more frequent update cycle, aiming to deliver continuous improvements and breakthrough capabilities to customers more quickly. This approach will ensure users remain at the forefront of AI advancements.