Open SourceHigh importance
K2-Horizon: six fully open models, from 0.9 billion to 375 billion parameters, with data, checkpoints and training code released
The K2 team published K2-Horizon, a fleet of six fully open language models, from a 0.9-billion-parameter dense model to a 375-billion MoE with 23 billion active, aimed at reasoning and agentic work in math, science, software engineering, research and office automation. According to the paper, the models lead or compete with the best open models of their size on agent and coding tests. The team introduces Mixture-of-Value Attention (MoVA) and diffusion adapters called Uno, which speed up generation by up to 2.2 times without a separate draft model. Following the LLM360 definition of open source, the release includes intermediate checkpoints from every stage, pre-training data (about 10 trillion synthetic tokens, in a 22-trillion corpus), training logs and all training, evaluation and infrastructure code.
Published in edition no. 36 of Noticiário do Joãozinho ·
Translated by Joãozinho AI from the original Brazilian Portuguese edition. Facts, numbers and sources are the same as in the original; quotes, video titles and source names stay in their original language. Read the original in Portuguese
Original source
Hugging Face PapersK2-Horizon artifacts
https://huggingface.co/papers/2611.21003