You'll noticeTier: how much it should matter to you
3Useful: my rating, 1 to 5
3Signal: worth watching, 1 to 5
ModelsArea: what it touches
ImprovementsKind: in v2.1.305,
What probably matters to youSection of the release
What
The context window is how much text a model can consider at once. Some models can reach 1 million tokens (1M).
The model schema accepts two new optional fields from remote config: maximum_window and clamp_long_context_without_beta.
Sonnet 4, Sonnet 4.6 and Opus 4.6 now carry maximum_window: 1e6 (1 million). Sonnet 4 also has clamp_long_context_without_beta set.
The capability record replaces has1mWithoutOptIn and unclampedButBilledPast200k with longContextWindowWithoutOptIn and apiContextWindowWithoutOptIn.
The check for whether a window counts as long-context without opt-in now compares it with the larger of those two values instead of testing a fixed size and a yes/no flag.
The long-context warning and the claude-desktop 1M notice now compare against a threshold instead of reading a yes/no value.
Why
This can change when a model is treated as having more than 200k context without opting in, which affects context limits, compaction and the 1M notice. The limits can now be adjusted from the server.
