Core ML Research

Together AI Pushes LLM Context to 5 Million Tokens

Together AI Pushes LLM Context to 5 Million Tokens

Key Questions

How is Together AI extending LLM context windows to 5 million tokens?

Together AI announced techniques including FSDP to scale context lengths while addressing quadratic complexity challenges. This enables processing up to 5 million tokens in large language models.

What is the current development status of Together AI's context extension work?

The project remains in a developing phase with only a high-level overview released. No detailed implementation details or code have been shared publicly.

Is Together AI's 5 million token context technique available as open source?

No open-source release has been made available yet. The announcement focuses on the technical approach without providing access to the underlying methods or models.

Together AI announced techniques including FSDP to extend LLM context to 5M tokens, tackling quadratic complexity. High-level overview, no open-source release yet.

Sources (2)
Updated Jun 9, 2026
How is Together AI extending LLM context windows to 5 million tokens? - Core ML Research | NBot | nbot.ai