The post has been translated automatically. Original language: Russian
Happy Programmer's Day, Astana Hub community and all code masters! 🎉
Today is your day, the day of those who know how to see structure in chaos, turn abstract tasks into working solutions and make the impossible possible. Every line of your code is a building block in the foundation of the digital economy, and it is precisely such efforts that move progress forward.
Special recognition to the Astana Hub participants: you are not just developing your own projects, you are creating an environment where innovations are born and new IT leaders are growing. Let your ideas resonate, your products attract users, and your ambitions attract resources for implementation.
We wish that the architecture of the systems is always well thought out, and refactoring is easy and stress-free; that tests confirm the reliability of solutions, and deployment remains calm and predictable. May there always be like—minded people around - those with whom even the most difficult tasks are solved in an atmosphere of support and true professional drive.
Let logic work flawlessly not only in the code, but also in achieving personal goals — to new heights, large-scale projects and bright victories!
Happy holidays!
Заметил, что почти никто не шумит, а новость стоит внимания: университет в Абу-Даби выкатил шесть ИИ-моделей — от 0,9 до 375 млрд параметров. Семейство K2 Horizon. Обучены с нуля, и всё открыто: веса, датасеты, рецепты обучения, код. Не просто дропнули модельку — выложили весь стек.
До GPT не дотягивают, но там есть модель на 7 млрд, которая берёт 70,6 на SWE-bench. Qwen3.5-9B на том же бенчмарке — 50,8. Разница серьёзная, особенно учитывая размер.
По-моему, главный поинт не в бенчмарках, а в том, что они открыли вообще всё. Обычно показывают финальную модель и всё. А тут — бери, разбирай, повторяй, улучшай. Редкий ход.
Share
