The o1 series introduced inference with internal “chain of thought” optimized by RL, excelling at math and code.
In September 2024, OpenAI launched o1-preview/o1: models that spend more test-time compute reasoning before answering, with strong gains on math, science, and programming contests.
The focus shifted from only “train bigger” to “think more at inference.” Competitors explored related ideas.
o1 foreshadowed the reasoning-system line that in 2026 would connect to announced advances on hard math problems.
OpenAI launched GPT-4o with low-latency voice and native vision, pushing multimodal conversational UI.
DeepSeek released competitive reasoning models with reportedly low training costs, altering market expectations.
Models optimized to think longer at inference defined 2025, with leaps in math contests and coding agents.