The AI agent startup said its 28 billion-parameter model sustained roughly 95 points in RULER long-context tests up to 4 million tokens, far above the roughly 1 million-token ceiling common among mainstream models.
verifying reliability
No specialized terms available for this topic.