Anthropic is enhancing Claude Sonnet 4 by introducing a 1 million token context window on its API. This represents a significant increase, a 5x jump from the previous 200K limit. The feature is now available as a public beta on Anthropic’s API and Amazon Bedrock, with plans for availability on Google Cloud’s Vertex AI soon. Initially, access is targeted at customers on higher rate tiers, with a broader rollout expected in the coming weeks.
Who is this for? Developers and enterprises handling heavy workloads.
What can it do? It can load entire repositories with more than 75,000 lines, digest extensive research or contract bundles, and maintain coherence across hundreds of tool calls without losing context.
Claude Sonnet 4 now supports 1 million tokens of context on the Anthropic API—a 5x increase.
— Claude (@claudeai) August 12, 2025
Process over 75,000 lines of code or hundreds of documents in a single request. pic.twitter.com/4hzltn9IoD
How it works and what it costs: Prompts up to 200K tokens maintain standard pricing. Inputs beyond 200K tokens are billed at a higher per-million-token rate, and outputs beyond 200K tokens are priced accordingly. Prompt caching helps reduce latency and costs, while batch processing can cut expenses by half for queued jobs.
Anthropic is positioning Sonnet 4 as the workhorse of the Claude 4 family, alongside Opus 4.1 and Haiku 3.5. This long-context capability aligns with its focus on agentic development, Claude Code, and enterprise deployment across AWS and Google Cloud channels. The aim is to establish Claude as the default choice for large-scale code and document reasoning.
