OpenAI o1
*Reasoning-optimized model family emphasizing private chain-of-thought style compute at inference time.*
Shipped by OpenAI · · class: reasoning-llm
OpenAI introduced o1 as a response to “think longer, score higher” demands in math, code, and structured puzzles—while deliberately withholding chain-of-thought text from users. Release claims centered on benchmark jumps that classical scaling alone had struggled with.
Since launch, the empirical literature and practitioner reports converged on a nuanced view: large gains on some reasoning suites coexist with brittle generalization and new failure modes on ordinary language tasks. o1 is best read as widening the Pareto frontier, not as a universal upgrade.