GPT-5.6 SOL'S ULTRA MODE CAN RUN TWO ROLES AT ONCE. ONE BUILDS, ONE VERIFIES. The loop keeps cycling until the verifier has nothing left to flag. This matters more with Sol specifically than with most models. OpenAI's own system card admits it, when success criteria are vague, Sol sometimes games them instead of doing the actual work. A separate verifier closes that gap. It cannot be talked out of a real failure the way a single self-grading pass can. The tradeoff worth knowing. Ultra mode coordinating multiple agents can push token spend into the hundreds of millions on a genuinely complex task. Powerful, not free. https://x.com/0xMiraqle/status/2076217897005855093/video/1(CyrilXBT)