Anthropic Introduces Claude 4 with New Advanced Programming Models

Anthropic has launched Claude 4, its new generation of artificial intelligence models specialized in programming and complex tasks. The family includes Claude Opus 4, designed for long-duration projects, and Claude Sonnet 4, a substantial improvement over the previous 3.7 model that maintains a balance between performance and efficiency.

Claude Opus 4 has achieved 72.5% on SWE-bench, a standard evaluation for programming models, and 43.2% on Terminal-bench. Meanwhile, Claude Sonnet 4 reaches 72.7% on SWE-bench, according to official Anthropic data. These scores position both models among the most competitive in the current market.

The new models incorporate an "extended thinking" system that allows combining internal reasoning with the use of external tools such as web searches. This functionality enables the models to alternate between analysis and resource consultation to improve their responses. Additionally, they can execute multiple tools in parallel and create memory files when they have access to local systems.

Several technology companies have begun integrating the models after internal evaluations. GitHub will incorporate Claude Sonnet 4 into its GitHub Copilot tool, while Cursor has reported improvements in complex code understanding. Replit highlights advances in precision for changes affecting multiple files, and Block indicates quality improvements during editing and debugging.

Anthropic has also confirmed the general availability of Claude Code, its development tool that integrates with editors like VS Code and JetBrains. The new beta extensions show proposed modifications directly in editor files.

The models are available in Claude's paid plans, with Claude Sonnet 4 also accessible in the free version. Prices remain at $15/$75 per million tokens for Opus 4 and $3/$15 for Sonnet 4, available through Anthropic's API, Amazon Bedrock, and Google Cloud Vertex AI.