Claude Sonnet 4 can process entire codebases with 1 million tokens

Anthropic has announced that its new Claude 4 Sonnet model can process entire codebases with more than 75,000 lines of code or research papers in a single request, thanks to the new 1 million tokens of context. This is a huge win for those who like to run complex data-driven programs with Claude.

Users can now load source files, tests, and documents with the model. Once uploaded, the AI will explore the project, identify improvements, and share them with the user to remove what’s lacking. You can now see extended document support thanks to the synthesis process, which sets legal contracts, research papers, or technical specifications.

Developers can now build agents that keep context across hundreds of tool calls and multi-step workshops. To provide developer support, Anthropic has launched API documentation and interaction histories.

This new context availability is now under beta access and available exclusively for API and Amazon Bedrock, and an expansion is coming to the Google Cloud Vertext AI platform. More importantly, Claude can analyze the relationship between documents without retouching the context.

One of the major benefits of this large context limit is that customers can use 1 million tokens with batch processing for up to 50 percent cost savings. It is confirmed that the prompts over 200,000 tokens will automatically get a price adjustment. That means the price for a million tokens input will cost $3, and $15 for output for less than 200k tokens.

On the other hand, the price for tokens exceeding 200k will be $6 per million tokens input and $22.50 per million tokens output.

(source)