Anthropic announces that Claude Sonnet 4 now supports up to 1M tokens of context — a leap that opens practical possibilities: analyzing huge projects, reviewing piles of papers, or keeping agents with long memory in a single request. What does that mean for your daily workflow and for development teams? (anthropic.com)
Qué cambia con 1M tokens
The change isn’t just a big number: Anthropic says it’s a 5x increase over the previous window, enough to process code with more than 75,000 lines or dozens of research articles in one call. That turns fragmented tasks — opening files, summarizing, refeeding context — into coherent, continuous flows. (anthropic.com)
Think of it like going from reading excerpts of a book to having the whole book on the table: the model can understand software architecture, relationships between files, or the throughline across hundreds of documents without losing track.
Casos de uso concretos
-
Large-scale code analysis: load entire repositories with tests and docs to detect cross-file dependencies and suggest refactors that consider the complete picture. (anthropic.com)
-
Document synthesis: summarize or compare contracts, technical specs, or collections of papers while keeping coherence between distant sections. (anthropic.com)
-
Agentes con contexto largo: build agents that retain interaction history, tool definitions, and API docs across hundreds of calls. Ideal for complex workflows and autonomous engineering sessions. (anthropic.com)
Precio y consideraciones de costo
To handle the higher compute cost, Anthropic adjusts pricing starting at 200K tokens of prompt. The official note shows differentiated rates for input and output in two tiers: up to 200K and above 200K tokens — check the pricing table for exact numbers by input/output and to optimize your prompt design. (anthropic.com)
A common tactic to reduce bill and latency is combining prompt caching and batch processing, strategies Anthropic also mentions as compatible with the long window. (anthropic.com)
Disponibilidad y cómo probarlo
Long-context support for Sonnet 4 is in public beta on Anthropic’s API and already appears on Amazon Bedrock, with Google Cloud Vertex AI promised soon. Early access targets Tier 4 customers and custom limits, and the company says availability will expand in the coming weeks. (anthropic.com)
If you want to experiment now, check the documentation and the pricing page to plan tests and budget. (anthropic.com)
Qué dicen los usuarios en producción
Anthropic shares customer cases: Bolt.new highlights improvements in code-generation workflows; iGent AI says the 1M tokens window has empowered engineering agents capable of multi-day sessions on real code. Those testimonials suggest that for teams already relying on models in production, the larger window helps move from experiments to scaled workflows. (anthropic.com)
"Claude Sonnet 4 remains our go-to model for code generation workflows... With the 1M context window, developers can now work on significantly larger projects," says a spokesperson quoted in the official note. (anthropic.com)
¿Y tú qué haces con esto?
If you’re a developer or tech lead: evaluate which parts of your pipeline need global context (architecture, tests, docs) and try prompt caching to control costs. If you’re a researcher or legal professional: think about grouping documents by topic and asking for comparative analyses in one pass.
It’s not magic: long windows mean higher cost per request and more demands on prompt design and data security. Is it worth it? That depends on whether your problem benefits from keeping everything in memory instead of fragmenting it.
If you want, I can help design a proof of concept: choose a repo or document set, decide how to split tokens between prompt and response, and define metrics to decide whether to migrate processes to the new long window.