Anthropic introduces Claude 4, the more powerful and durable AI

Maxime Hiez
Anthropic
27 Jun, 2025

Introduction

In an artificial intelligence market dominated by OpenAI, Google, and Microsoft, Anthropic continues to forge its own path. With the launch of the Claude 4 family, the Californian startup founded by former OpenAI researchers is affirming its ambitions : to offer AI that is more powerful, more secure, and more aligned with the needs of professional users.

A new generation of models

The Claude 4 range consists of two main models :

Claude 4 Opus : the most advanced model, reserved for paid subscribers
Claude 4 Sonnet : a free, yet surprisingly powerful version

These models succeed Claude 3.5 and incorporate major improvements in reasoning, coding, contextual understanding and security.

Multi-stage reasoning and cognitive endurance

Claude 4 Opus is capable of maintaining logical consistency over long sessions, with the ability to reason across multiple stages in complex workflows. In internal tests, it coded autonomously for several hours, achieving a score of 72.5% on the SWE-bench benchmark, compared to 54.6% for GPT-4.1.

This makes it one of the most powerful models on the market for advanced programming tasks, technical problem solving or strategic planning.

Claude 4 Sonnet : The free version that surprises

Remarkably, Claude 4 Sonnet, despite being the free version, outperformed Opus in several coding tests conducted by ZDNet. It passed 100% of the tests, compared to 50% for Opus in the same environment.

This demonstrates that Sonnet is not a limited version, but rather a robust model, capable of competing with the best AI on the market, even in demanding contexts such as software development.

A high-level development assistant

Claude 4 is particularly effective in development environments :

Intelligent code completion
Refactoring and documentation
Plugin creation (e.g., WordPress)
Unit testing and bug fixing

It is integrated into Claude Code, an AI-assisted development environment, a direct competitor to GitHub Copilot, Cursor, or Gemini Code Assist.

Security, alignment, and reward hacking reduction

Anthropic has focused on security and behavioral alignment. Claude 4 is less prone to reward hacking, a common problem in language models. It follows instructions better, produces more reliable responses, and is more respectful of user-imposed limits.

This makes it a strategic choice for companies concerned with compliance, ethics and control.

Accessibility, integration, and pricing

Claude 4 Sonnet is accessible for free via the Claude web interface
Claude 4 Opus is available via a $20/month subscription or via the Amazon Bedrock and Google Vertex AI platforms, with pay-as-you-go pricing ($15 / $75 per million input / output tokens)

Both models are designed for easy integration into enterprise workflows, with robust APIs and clear documentation.

Note : Prices in USD before applicable taxes.

A deliberate ambition

Anthropic is targeting revenue of $12 billion by 2027, compared to an estimated $2.2 billion for 2025. To support this growth, the company has raised several billion dollars, notably from Amazon and Google.

With Claude 4, it is no longer content to keep pace: it is setting its own pace in the race for generative AI.

Conclusion

With Claude, Anthropic 4 marks a decisive step in the evolution of AI models. More robust, more reliable, and more accessible, it stands out as a serious alternative to GPT-4 and Gemini, particularly for developers, businesses, and researchers. By combining performance, security, and accessibility, Anthropic confirms its position as a key player in the AI of tomorrow.