Anthropic Cuts Costs for Giant Context Windows

Anthropic Reduces Costs for One Million Token Context Windows

Anthropic recently announced a significant reduction in costs associated with the use of extra-large context windows in its artificial intelligence models. The Opus 4.6 and Sonnet 4.6 models can now process up to one million tokens at no additional charge, a major advancement compared to the 100% surcharge that was previously applied for requests exceeding 200,000 tokens.

Pricing and New Limits

Despite the removal of the surcharge, the base prices of the models remain unchanged: Opus 4.6 is priced at $5/$25 per million tokens for input and output, while Sonnet 4.6 is at $3/$15. Now, whether the prompt contains 9,000 or 900,000 tokens, the cost remains the same. Additionally, the media limit per request has been increased from 100 to 600 images or PDF pages, thereby expanding usage possibilities.

Availability and Performance

This new pricing is applicable to users of Claude Code in its Max, Team, and Enterprise versions, and is accessible through platforms such as Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry (with the exception of the media limit).

In terms of performance, the Opus 4.6 model has demonstrated the ability to maintain stable performance even at full context length, according to the GraphWalks BFS benchmark, which evaluates logical reasoning over large amounts of text. Anthropic claims that its models achieve maximum accuracy among comparable models during benchmark testing, although the challenge of declining accuracy with increasing context length remains a problem to be solved.