Model techniques map
Techniquesmodel architecturecontext capacity

implementation detail · filed under model architecture

1M-token context window

A context capacity of up to one million tokens, distinguished from the separately mentioned 256K context window.

Also called 1M context length, 1M context, 1M context window, long context window, One-million-token context window.

sources
4
models
5
labs adopt it
2
strongest
core

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 2default 1core 1

Documented in

Evidence

4 spans quoted from the sources, strongest treatment first.

1M-token context window, suited for mixed-modality input and multi-agent collaboration

coremodel architecturein MiMo-V2.6-FlashXiaomi

1M context is now the default across all official DeepSeek services

defaultotherin DeepSeek-V4DeepSeek

Its 1M context window supports complete documents, extended conversations, and complex task contexts in a single pass

usedmodel architecturein MiMo-V2.5Xiaomi

supports up to 1 million tokens of context

usedmodel architecturein MiMo-V2.5Xiaomi

Filed alongside

Other methods under model architecture :: context capacity.