GPT-6 Pro Feels Like an Oracle, But the Ultra Mode in Codex Does Something Entirely Different
emollick · x · 2026-11-01
Ethan Mollick says the GPT-6 Pro series excels at one-shot problems involving complex calculations, references, or sophisticated ideas — slow, not pleasant, but with an "ask one question of the great oracle" feel. He flags a confusion point: Pro mode is only in the Chat interface; the highest thinking tier in Codex and ChatGPT Work is called Ultra and spawns many agents to solve problems in parallel — something entirely different.
Related event: Ethan Mollick Reviews GPT-6 Pro as an Oracle(3 posts)→
More from Models
- Memdoor claims 5-model Qwen team beats flagship LLMs on LiveCodeBench Hard — PferdOne · 2026-11-01
- Claude Pro user burns 40% of weekly quota in one day, says $500 tier now needed — lxfater · 2026-11-01
- Why does OpenAI own the entire price-performance Pareto frontier? — mrvoidance · 2026-11-01
- GPT-6.5 'Bel' may be a bigger jump than expected, hinges on pretraining rumors — VraserX · 2026-11-01
- Anthropic rolls out monthly Claude API credits up to $500 for Max and Team plans — samuelcolvin · 2026-11-01
- Bonsai 2 27B's Chain-of-Thought Sounds Eerily Human, Reddit Notes — GrokiniGPT · 2026-11-01