That’s up five-fold from prior limits. This larger context window can be accessed directly from Anthropic through its application programming interface (API) or Amazon Bedrock.
The extended context window allows Claude to process far larger data sets in a single prompt, such as entire codebases of more than 75,000 lines or dozens of research papers, according to a Tuesday (Aug. 12) blog post.
Anthropic said the change enables new classes of data-intensive applications in software engineering, research and automation.
For developers, the company pointed to three key use cases.
First: large-scale code analysis, where Claude can load all source files, tests and documentation to map project architecture and suggest improvements that account for the entire system design.
Advertisement: Scroll to Continue
Second: document synthesis, which allows users to process and analyze large volumes of legal, academic or technical material while preserving context across hundreds of documents.
Third: context-aware agents, where developers can maintain full coherence across hundreds of tool calls and multi-step workflows by including complete API documentation, tool definitions and interaction histories.
Other AI models with large context windows include Google’s Gemini, OpenAI’s GPT-4.1 and Alibaba’s Qwen, with each offering one-million-token context windows. Google has promised to double the context window, while Meta has exceeded its rivals with a 10-million window for Llama 4 Scout.
These and other developments from tech companies are the main ones driving AI adoption, according to a PYMNTS Intelligence report, Tech on Tech: How the Technology Sector is Powering Agentic AI Adoption.
“The technology has been hyped ever since OpenAI debuted in January an autonomous AI agent that can fill out online forms, make grocery lists and generate memes,” according to the report. But a sticking point has been the size of the input capacity.
Claude Sonnet 4’s expanded capacity comes with a revised pricing structure. Prompts up to 200,000 tokens cost $3 per million tokens for input and $15 for output. Prompts above that threshold are priced at $6 for input and $22.50 for output.
Anthropic noted that prompt caching, or storing data that’s already been processed to avoid reprocessing, can reduce costs and latency, while batch processing with the one-million context window can deliver an additional 50% cost savings.
Early adopters are already integrating the expanded model into production workflows.
Bolt.new, which offers a browser-based web development platform, uses Claude Sonnet 4 for code generation. London-based iGent AI is applying the feature to Maestro, its AI software engineering agent that turns conversations into executable code.
Long-context support for Claude Sonnet 4 is currently limited to customers with Tier 4 and custom rate limits, with wider rollout planned in the coming weeks. Anthropic said it is also exploring ways to extend the feature to other Claude products.
