Verified65% confidenceFactExact time
OpenAI's o1 model was released in September 2024; through reinforcement learning it was trained to perform internal reasoning before outputting a final answer, with intermediate reasoning steps existing as special tokens that are invisible to users.
3
Sources
65%
Confidence
Long-term
Relevance
7/21/2026
First Seen
Sources
Related Claims
UnverifiedOpenAI的o系列模型（o1、o3）是强化学习推理（RLR）路线的代表性产物，以可验证的客观标准作为奖励信号77% similarVerifiedThe OpenAI o1 series develops Chain-of-Thought capabilities through reinforcement learning training, far surpassing traditional GPT-4 in mathematical proofs, algorithm design, and complex logical reasoning.75% similarVerifiedOpenAI's o1 series represents a new paradigm called "test-time compute"74% similarUnverifiedOpenAI launched the Deep Research feature based on the o3 model in early 2025, capable of completing in minutes the literature review work that would take human researchers hours.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/575742API
curl https://kongchang.com/api/v1/knowledge/claims/575742MCP
get_claim(id=575742)