Artificial intelligence startup Anthropic today announced that its popular Claude Sonnet 4 LLM model now supports up to 1 million contextual tokens. Following Google Gemini, Anthropic becomes the second major model provider to deliver 1 million contextual tokens.
Previously, the Claude Sonnet 4 API only supported 200,000 contextual tokens. With this five-fold expansion, developers can send their entire code base of more than 75,000 lines of code in a single request.
Extended context support is now in public testing on the Anthropic API and Amazon Bedrock, with Google Cloud Vertex AI coming soon. However, long context support is currently limited to Tier 4 developers and requires custom rate limits. Anthropic notes that this feature will be made available to more developers in the coming weeks.
Since larger token windows require higher computing power, Anthropic has introduced special pricing. For prompts under 200,000 tokens, Sonnet 4 costs $3 per million input tokens and $15 per million output tokens. For prompts with more than 200,000 tokens, the cost is $6 per million input tokens and $22.50 per million output tokens.
Developers can reduce costs by using fast caching and batch processing. For example, batch processing can offer a 50% discount on 1M context window pricing.
In a recent Reddit AMA session, OpenAI leaders discussed supporting long context windows for their models. OpenAI CEO Sam Altman said that OpenAI hasn't seen strong user demand for long context lengths, but if there is enough interest, they are willing to support it. With limited computing power, they want to focus on other priorities.
Michelle Pokrass from the OpenAI team wrote that they would have liked to provide longer context in GPT-5, up to 1 million tokens, especially for API use cases, but they did not do so due to high GPU demand.
Anthropic's 1M context support puts it in direct competition with Google Gemini for long context capabilities, putting pressure on OpenAI to rethink its roadmap.
