Researcher: GPT 5.6 Sol Ultra Beats Pro for Long-Horizon Hard Problems

arankomatsuzaki · x · 2026-08-24

Researcher Ara Kometzaki responded to Mikhail Parakhin's weekly reminder that OpenAI should pay more attention to the Pro (best-of-n) configuration, saying that gpt 5.6 sol ultra on the ChatGPT app is similar to Pro but can work even longer and transparently shows all intermediate steps — except for compaction summaries and chain-of-thought.

He always picks Ultra over Pro for hard problems.

Context: Parakhin argues Pro (best-of-n) is by far the best "model" for math/ML/deep discussions, yet it isn't even available in the ChatGPT app.

Related event: Ex-Microsoft CTO Champions ChatGPT Pro as Tests Show Ultra Overtakes(2 posts)→

Original post →

More from Models

Models channel →