Anthropic Announces Claude Opus 4, the World's Most Powerful Programming Model

Anthropic today announced the next generation of its Claude models: Claude Opus 4 and Claude Sonnet 4. The company claims that Claude Opus 4 — the most advanced version — is the world's most powerful programming model , excelling at agentic workflows and handling complex, long-running tasks. Meanwhile, Claude Sonnet 4 improves programming and logic performance over the previous Claude Sonnet 3.7 version.

 

According to the comparison chart below, Claude Sonnet 4 scored an all-time high of 72.7% on SWE-bench Verified — a popular benchmark for automated programming capabilities. With this strength, Microsoft's GitHub announced that Sonnet 4 will be the core model for the new programming agent in GitHub Copilot .

Anthropic describes the performance of the new models as follows:

These models advance customers' AI strategies across the board: Opus 4 breaks boundaries in programming, research, writing, and scientific discovery, while Sonnet 4 brings cutting-edge performance to everyday tasks as an immediate upgrade from Sonnet 3.7.

 

Both new Claude 4 models support two modes:

1. Almost instant response
2. Extended Thinking (for problems that require in-depth analysis)
   Programmers can switch between the two modes as needed.

Claude free users only get access to the Sonnet 4 model, while Pro, Max, Team, and Enterprise users get access to both models and extended thinking. Anthropic also says the models are available on the Anthropic API , Amazon Bedrock , and Google Cloud's Vertex AI . Notably, Anthropic isn't offering discounts on the new models:

1. Claude Opus 4: $15/1 million input tokens and $75/1 million output tokens
2. Claude Sonnet 4: $3/1 million input tokens and $15/1 million output tokens

Along with the new models, Anthropic announced updates to Claude Code with support for background tasks via GitHub Actions , native integration with VS Code and JetBrains , and more. Additionally, the Anthropic API adds a code execution tool, MCP connectivity, Files API, and the ability to cache prompts for up to 1 hour .