Anthropic just released Claude Sonnet 4.6, and it is a significant leap. The model ships with a 1 million token context window, a new Adaptive Thinking engine, and benchmark results that put it ahead of every publicly available model.

1M Token Context Window

The headline feature is context length. Claude Sonnet 4.6 can process up to 1 million tokens in a single request. That is roughly 750,000 words, or about 10 full-length novels. For developers, this means you can feed entire codebases, documentation sets, or long conversation histories into a single prompt without chunking or summarization.

In practice, this changes how you architect AI-powered applications. Retrieval-augmented generation (RAG) pipelines that previously required careful chunking strategies can now pass much larger context blocks directly. For code analysis, you can include hundreds of files in a single request and ask the model to reason across all of them.

Adaptive Thinking Engine

The second major feature is Adaptive Thinking, a system that dynamically allocates compute based on the complexity of each part of a prompt. Simple questions get fast, efficient responses. Complex reasoning tasks trigger deeper analysis automatically, without the user needing to specify a "thinking mode" or adjust parameters.

This is different from the extended thinking features in earlier Claude models, which required explicit opt-in. Adaptive Thinking is always on and makes its own decisions about how much computation to dedicate to each response. The result is faster responses for straightforward tasks and more thorough analysis for hard problems, all within the same API call.

Benchmarks

The numbers are impressive. Claude Sonnet 4.6 scores 72.5% on OSWorld, a benchmark for real-world computer use that tests an AI's ability to complete tasks across operating systems. It also shows strong results on SWE-bench, MMLU, and HumanEval, though Anthropic has been characteristically measured in how they present these figures.

What matters more than any single benchmark is the consistency. Sonnet 4.6 performs well across coding, analysis, writing, and multimodal tasks without the tradeoffs that typically come from optimizing for one category.

Pricing

Claude Sonnet 4.6 is priced at $3 per million input tokens and $15 per million output tokens. This is competitive with other frontier models and represents a significant capability-per-dollar improvement over the previous Sonnet generation. The 1M context window is available at the same per-token rate, with no premium for longer contexts.

What This Means for Developers

For teams building AI-powered products, the larger context window reduces the need for complex chunking and retrieval strategies. Applications that process long documents, large codebases, or extended conversation histories can now pass significantly more information in a single request. The Adaptive Thinking engine removes the need to manually toggle reasoning modes, simplifying integration.

The combination of context length, adaptive compute, and strong cross-domain performance makes Claude Sonnet 4.6 a versatile foundation for a wide range of applications, from code analysis and document processing to creative tools and multimodal workflows.