A large multimodal reasoning model that thinks before it speaks — it works through visual and textual problems step by step before committing to an answer. With 235 billion total parameters but only 22 billion active at once, it balances depth with efficiency. It handles complex image understanding and multi-step reasoning well, though its size means deployment requires serious hardware planning.