And here we go again: a finetuned DeepSeek scientific model allegedly achieved 40,44% on HLE

→ View original post on X — @kimmonismus