- Anthropic released Opus 4.6 and Sonnet 4.6 with a 1M context window, roughly equivalent to 1,000–2,000 pages or 4–5 novels.
- Unlike earlier large-context models from Google and OpenAI, Anthropic's models maintain high recall at 1M tokens, avoiding 'context rot' where performance degrades.
- Longer context windows are crucial for agentic workflows, preventing the need for 'compaction' which loses detail and can lead to inefficiencies in complex tasks.
- Benefits include better handling of large document sets in fields like law and finance, and preserving subtle nuances from full conversations that summaries miss.
- Anthropic does not charge extra for the 1M context, unlike competitors who double prices past 200–272K tokens, and includes it in Max and Business subscriptions.