GTC-2.5 Preview is the first conversational foundation model released by TAI Research. Built upon the open‑source MiniMind project and trained on a V100 GPU, this 64M‑parameter model serves as a validation of our training pipeline and a baseline for subsequent larger models.

The model demonstrates foundational capabilities in dialogue, knowledge retrieval, and code generation. While it is a compact research preview, it establishes a reference point for future iterations and downstream applications, including domain‑specific assistants and lightweight edge deployments.

We extend our gratitude to the MiniMind team for their open‑source contributions.

📄 Download System Card (PDF)