Anthropic announced that its Claude Sonnet 4 model can now handle 1 million tokens of context in a single request — a fivefold increase over previous limits. The feature, available in public beta through Anthropic’s API and Amazon Bedrock, enables developers to load full codebases of more than 75,000 lines without breaking them into smaller parts.
This expanded capacity allows Claude to analyze complete projects, synthesize large document sets, and support AI agents that maintain coherence across complex workflows. Anthropic says the model achieved 100% accuracy in internal “needle in a haystack” tests, demonstrating its ability to locate details within massive text volumes.
Developers and platform partners report that the update supports true production-scale engineering, including multi-day coding sessions and repository-wide analysis. Early enterprise adopters span industries from financial services to legal tech.
Pricing has been adjusted for the larger context window. Inputs up to 200,000 tokens retain current rates of $3 per million input tokens and $15 per million output tokens, while larger prompts cost $6 and $22.50 respectively. Anthropic argues that caching and quality gains offset higher costs compared to retrieval-augmented approaches.
The launch comes as Anthropic faces competition from OpenAI and Google, both of which offer similar context sizes. However, Anthropic emphasizes Claude’s strength in coding and reasoning, and currently holds 42% of the AI code generation market, though much of its revenue relies on key customers like GitHub Copilot and Cursor.
The 1 million token window, equivalent to about 750,000 words, marks a technical leap in AI memory and attention mechanisms. While raising safety considerations, the upgrade positions Claude as a full-scale development partner capable of understanding and improving entire software systems.
