Moonshot AI has officially launched Kimi K3.1 Pro, a new mixture-of-experts model featuring 2.8 trillion total parameters and 104 billion active parameters. According to reporting by MarkTechPost, the architecture supports a one-million-token context window and achieves a 65.7 percent score on Terminal-Bench 4.0. API pricing is set at two dollars per million input tokens and ten dollars per million output tokens, with open weights scheduled for release on November 4, 2026.
Architecture and Scale
Large language models utilizing mixture-of-experts designs activate only a fraction of their total parameters for any given token, balancing massive capacity with computational efficiency. The combination of a 2.8-trillion parameter ceiling, a 104-active parameter routing mechanism, and a one-million-token context window places this release among the higher-capacity configurations currently emerging from the AI sector. The specific hardware requirements and training dataset details were not disclosed in the announcement.
Unknowns and Future Outlook
While the benchmark score and pricing structure provide initial performance and economic metrics, several details remain absent. The announcement did not specify the exact hardware infrastructure used to train or serve the model, nor did it outline the specific mixture-of-experts routing architecture employed. Observers will need to monitor how the model performs on broader evaluation suites ahead of the open weights release in late 2026.
Kimi K3.1 Pro represents Moonshot AI’s latest push into high-context, high-parameter artificial intelligence development.
Lead image: Christina Morillo / Pexels
