Anthropic launches 1M-token context window on Opus 4.6 and Sonnet 4.6

March 17, 2026 Anthropic has made a 1 million token context window generally available for Claude Opus 4.6 and Sonnet 4.6, allowing users to process large codebases, documents and multi-step workflows in a single request. The company is applying standard per-token pricing across the entire window, meaning a 900,000-token request costs the same per-token rate as a much smaller query.

The update removes a key limitation for developers working with large inputs. Standard pricing is set at $5 input and $25 output per million tokens for Opus 4.6, and $3 input and $15 output for Sonnet 4.6, with no additional cost for extended context usage.

A context window defines how much information a model can reference at once, including prompts, conversation history and uploaded data. At one million tokens – roughly 750,000 words – Claude can now operate across full codebases, lengthy contracts or extended agent workflows without requiring summarisation or context management techniques.

Anthropic also introduced full rate limits across all context sizes, removing previous throughput constraints tied to longer inputs. Requests now process at standard speeds regardless of whether the input is 50,000 or 950,000 tokens.

Additional updates include support for up to 600 images or PDF pages per request, up from 100, and removal of the beta header previously required for large-context queries. The expanded capabilities are available across Anthropic’s platform as well as integrations with Microsoft Azure and Google Cloud Vertex AI.

The company said Claude Opus 4.6 maintains performance at scale, scoring 78.3 per cent on the MRCR v2 benchmark at full context length. This addresses a common issue in large-context models, where retrieval accuracy can degrade as input size increases, a phenomenon often referred to as “context rot.”

For enterprise use cases, the change reduces the need for engineering workarounds such as chunking, summarisation or context pruning. Long-running AI agents can now retain full execution traces across multiple steps, while document-heavy workflows can be processed in a single pass.

The update also extends to Claude Code, where Max, Team and Enterprise users can now run sessions with the full 1M context window without additional allocation requirements.

Top Stories

Related Articles

March 17, 2026 Tencent Cloud has become an official sponsor of the OpenClaw open-source AI agent community, days after its more...

March 17, 2026 Dell has reduced its workforce to about 97,000 employees, cutting roughly 11,000 roles over the past year more...

March 17, 2026 OpenAI is refocusing its business on coding and enterprise products after internal leadership said its broad “do more...

March 16, 2026 OpenAI CEO Sam Altman said artificial intelligence may eventually be sold as a metered utility similar to more...

Jim Love

Jim is an author and podcast host with over 40 years in technology.

Share:
Facebook
Twitter
LinkedIn