{
  "id": 5446404,
  "title": "K2 Horizon: A connected fleet of six open models",
  "url": "https://urgent.news/2026/09/03/k2-horizon-a-connected-fleet-of-six-open-models",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-09-03T15:36:43.000Z",
  "source": {
    "name": "Hacker News Best",
    "slug": "hacker-news-best",
    "url": "https://ifm.ai/blog/k2/"
  },
  "original_language": "en",
  "account": "IFM has announced the release of K2 Horizon, a connected fleet of six open models: 375B-A23B, 36B-A4B, 32B, 7B, 3.7B, and 0.9B. These models exhibit top-tier performance across various tasks, including reasoning, mathematics, coding, agentic tasks, and general capabilities.\n\nThe 0.9B, 3.7B, and 7B models set new state-of-the-art results in their respective scales, while the 36B-A4B model achieves efficiency with its unique Mixture-of-Expert design. The 32B and 375B-A23B models rank among the top in their comparison classes.\n\nNotably, the smaller models in K2 Horizon demonstrate impressive capabilities. For instance, the 0.9B model achieves an AIME 2026 score above 48, while the 3.7B and 7B models excel in complex tasks such as software engineering and multi-step environments.\n\nThe release of K2 Horizon aims to provide a transparent and comprehensive model, offering both highly capable models and the training recipes used to create them. This allows researchers to study advanced capabilities and developers to reproduce, adapt, and extend the methods used in the models.\n\nWith a sparse MoE architecture, 375B-A23B is the largest and most capable model among the fleet, ranking among the top models for general, reasoning, coding, and agentic evaluations. The fleet's models occupy different performance sweet spots, balancing capability, adaptability, and local deployability.\n\nK2 Horizon is designed as a connected family, sharing core architectural decisions, training methodology, interfaces, evaluation infrastructure, and deployment tooling. This allows developers to easily move between sizes, while researchers can study capability across scale in a more controlled setting.",
  "summary": "Article URL: https://ifm.ai/blog/k2/ Comments URL: https://news.ycombinator.com/item?id=49551760 Points: 238 # Comments: 79",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}