Claude Sonnet 4 Expands to 1 Million Token Context
Created on August 12|Last edited on August 12
Comment
Anthropic has announced that Claude Sonnet 4 can now handle up to 1 million tokens of context, a fivefold increase from its previous limit. This expansion, now in public beta, allows the model to process entire codebases exceeding 75,000 lines or dozens of research papers in one request. The feature is currently available through the Anthropic API and Amazon Bedrock, with Google Cloud’s Vertex AI integration on the way.
Enabling larger and more complex workflows
The jump in context size means Sonnet 4 can now power scenarios that require retaining and reasoning over vast amounts of information without losing track of details. For software engineering, it enables complete project analysis across all files, tests, and documentation in a single pass. In document-heavy fields like law or academia, it can synthesize insights from hundreds of documents at once. It also benefits multi-step autonomous agents, allowing them to maintain memory across hundreds of tool calls and workflow stages.
Updated pricing structure for long context
Handling more than 200,000 tokens in a prompt incurs higher processing costs due to the added computational load. For input prompts under 200,000 tokens, pricing remains at $3 per million tokens for input and $15 for output. Above that threshold, input pricing doubles to $6 per million tokens, and output rises to $22.50. Anthropic notes that prompt caching and batch processing can offset these increases, with batch mode potentially saving up to 50% in costs.
Updated pricing structure for long context
Bolt.new, which integrates Claude into its web development platform, reports improved ability to work with larger projects while maintaining high accuracy. London-based iGent AI says the leap to 1 million tokens has transformed its Maestro AI agent, enabling sustained multi-day engineering sessions on real-world codebases. Both companies describe the change as a shift toward true production-scale AI workflows.
Availability and next steps
Long context for Sonnet 4 is rolling out first to Anthropic API customers with Tier 4 or custom rate limits. Wider access is planned in the coming weeks, alongside further availability in Amazon Bedrock and Google Cloud’s Vertex AI. Anthropic is also exploring extending long-context capabilities to additional Claude products.
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.