AISystem

Kimi K3

Moonshot AI's 2.8-trillion-parameter Mixture-of-Experts model with native vision, the largest open-weight model out of China and the successor its chief executive says K2 should help build.

open-weight-model

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Kimi K3 carries 2.8 trillion Mixture-of-Experts parameters on Delta Attention with native vision (quickstart); on arrival it took first in the Frontend Code Arena and dethroned Fable 5 in six of seven domains, the open crown changing hands in a day, then beat every closed rival on SpreadsheetBench 2. In this cluster it is the object of the roadmap its chief executive stated as we want K2 to help build K3, and the workhorse of the cybersecurity benchmark on which Fable 5 refused every task. Days later the White House alleged it had been distilled from Fable through a covert platform, and the full weights followed as the first open 3-trillion-class model.

Development timeline

Newest reporting dates first. Reporting dates may differ from event dates.

    Knowledge graphopen in browser →

    Developed by
    Evaluated on

    Cite or correct this record

    Suggest a correction — opens a GitHub draft for you to review and submit.