Trying out QvQ—Qwen’s new visual reasoning model