Anthropic celebrated a major breakthrough in the field of artificial intelligence on August 13th, officially announcing that its Claude Sonnet 4 model now supports millions of contextualized words, a fivefold increase from the previous limit of 200,000 words. This upgrade enables developers to process over 75,000 lines of code or dozens of academic papers in a single pass, significantly improving the efficiency of complex tasks. This feature is currently available for testing in the Anthropic API and Amazon Bedrock, with Google Cloud Vertex AI soon to follow. Notably, this feature is initially only available to Tier 4 developers and comes with customizable rate limits, but the company promises to gradually expand its availability over the coming weeks.
To address surging computing demand, Anthropic has introduced a tiered pricing strategy: requests under 200,000 words remain priced at the same price ($3 per million words for input, $15 per million words for output), while requests with longer contextualized words double in price, but can achieve a 50% cost savings through batch processing. This strategy ensures a basic user experience while providing flexible options for professional applications. Interestingly, OpenAI admitted in a recent Reddit AMA that while it had planned to introduce million-word support for GPT-5, it had been shelved due to GPU resource constraints. CEO Sam Altman even stated that they "did not see strong market demand."
This upgrade directly challenges Google Gemini's advantage in long-text processing, forcing the industry to reevaluate its technology approach. With companies like Bolt.new successfully implementing million-word support in their code analysis platforms, Claude Sonnet 4 is driving the transformation of AI from fragmented interactions to systematic collaboration.
