Artificial intelligence startup Anthropic announced today that their popular Claude Sonnet4 language model (LLM) now supports up to 1 million context tokens. This update makes Anthropic the second major model provider, after Google Gemini, to offer such long context support.
Previously, the context limit for the Claude Sonnet4 API was 200,000 tokens, and this expansion increases it fivefold, allowing developers to submit entire codebases of over 75,000 lines of code in a single request. This change will greatly benefit developers, especially when dealing with complex tasks or large-scale data.
The long context support for Claude Sonnet4 is now in public testing on Anthropic's API and Amazon Bedrock, and Google Cloud Vertex AI is also set to launch similar features soon. However, the current long context support is only available to Tier4 developers and comes with custom rate limits. Anthropic stated that this feature will be gradually opened to more developers in the coming weeks.
With the expanded context window comes higher computational demands. To address this, Anthropic has introduced a new pricing strategy: for prompts under 200,000 tokens, the cost of Sonnet4 is $3 per million input tokens and $15 per million output tokens. For prompts exceeding 200,000 tokens, the cost increases to $6 per million input tokens and $22.5 per million output tokens. Notably, developers can effectively reduce costs by using fast caching and batch processing, with batch processing even offering a 50% discount on the pricing for a 1M context window.
Recently, during a Q&A session on Reddit, the leadership of OpenAI discussed plans to support long context windows. OpenAI's CEO Sam Altman stated that although they have not yet observed strong user demand for long context, they would consider supporting this feature if there is sufficient interest. Michelle Pokrass from the OpenAI team mentioned that they originally planned to provide up to 1 million token context support in GPT-5, particularly for API use cases, but were unable to do so due to high GPU requirements.
With Anthropic launching support for 1M context, it directly competes with Google Gemini in long context capabilities, undoubtedly putting pressure on OpenAI to re-evaluate its product roadmap.