one year on
Anthropic expands Claude Sonnet 4 context window to 1 million tokens
A 5x jump to 1M tokens lets developers feed entire codebases or dozens of research papers into a single prompt, with pricing stepping up for prompts over 200K tokens.
Anthropic today expands Claude Sonnet 4 to support up to 1 million tokens of context on the Anthropic API, a fivefold increase from the previous 200,000-token limit. The larger window is in public beta for customers with Tier 4 or custom rate limits, and is also available on Amazon Bedrock, with Google Cloud’s Vertex AI support coming soon.
At 1 million tokens, Claude can ingest entire codebases exceeding 75,000 lines of code, dozens of research papers, or extensive document sets in a single request. Anthropic pitches the expanded context for large-scale code analysis, multi-document synthesis, and building agents that stay coherent across hundreds of tool calls.
Pricing steps up for prompts that exceed 200,000 tokens: input doubles to $6 per million tokens and output rises to $22.50 per million, against $3 and $15 for shorter prompts. Anthropic notes that prompt caching can reduce latency and costs, and that batch processing offers an additional 50 percent cost savings.
Early customers include Bolt.new, whose CEO Eric Simons says the model outperforms others in production for code generation, and iGent AI’s Maestro agent, which CEO Sean Ward says now enables “multi-day sessions on real-world codebases.” The upgrade caps a busy stretch: Anthropic shipped Claude Sonnet 4.1 a week ago.
The record
CEO of Bolt.new said Claude Sonnet 4 with the larger context window lets developers work on significantly larger projects while maintaining high accuracy in production.
CEO of iGent AI said the 1M-token context empowers autonomous multi-day sessions on real-world codebases.
One year later — open only if you can handle spoilers
The 1M window left beta over the following weeks — Google's Vertex AI went live on August 26, joining Amazon Bedrock — and long context slid from differentiator to table stakes, with Google's Gemini already advertising comparable limits. Anthropic carried the feature into its newer models; by February 2026 the million-token window shipped at standard pricing on Opus 4.6 and Sonnet 4.6, retiring the premium surcharge that debuts here. The wager that huge prompts would displace retrieval pipelines proved half right: RAG stuck around, but "just paste the whole codebase" quietly became a normal way to work.
The Weekly Replay · free by email
This week, one year ago — every Sunday.
One email each Sunday: the week's replayed AI news, with the one-year-later annotations included. Written like it's breaking — dated like it isn't.
Free · double opt-in · unsubscribe anytime · privacy