OpenAI's o1

Switch to

- [ChatGPT(opens in a new window)](https://chatgpt.com/?openaicom-did=ad31255f-bbda-4d1c-bfb3-774766dcd5d9&openaicom_referred=true)
- [Sora(opens in a new window)](https://sora.com/)
- [API Platform(opens in a new window)](https://platform.openai.com/)

Learning to reason with LLMs \| OpenAI

September 12, 2024

[Release](https://openai.com/research/index/release/)

# Learning to reason with LLMs

[Contributions](https://openai.com/openai-o1-contributions/) [Use o1(opens in a new window)](https://chatgpt.com/?openaicom-did=ad31255f-bbda-4d1c-bfb3-774766dcd5d9&openaicom_referred=true)

Listen to article

Share

OpenAI o1 ranks in the 89th percentile on competitive programming questions (Codeforces), places among the top 500 students in the US in a qualifier for the USA Math Olympiad (AIME), and exceeds human PhD-level accuracy on a benchmark of physics, biology, and chemistry problems (GPQA). While the work needed to make this new model as easy to use as current models is still ongoing, we are releasing an early version of this model, OpenAI o1‑preview, for immediate use in ChatGPT and to [trusted API users⁠(opens in a new window)](https://platform.openai.com/docs/guides/rate-limits/usage-tiers).

Our large-scale reinforcement learning algorithm teaches the model how to think productively using its chain of thought in a highly data-efficient training process. We have found that the performance of o1 consistently improves with more reinforcement learning (train-time compute) and with more time spent thinking (test-time compute). The constraints on scaling this approach differ substantially from those of LLM pretraining, and we are continuing to investigate them.

![The image shows two scatter plots comparing "o1 AIME accuracy" during training and at test time. Both charts have "pass@1 accuracy" on the y-axis and compute (log scale) on the x-axis. The dots indicate increasing accuracy with more compute time.](https://images.ctfassets.net/kftzwdyauwt9/3OO9wpK8pjcdemjd7g50xk/5ec2cc9d11f008cd754e8cefbc1c99f5/compute.png?w=3840&q=90&fm=webp)

o1 performance smoothly improves with both train-time and test-time compute

## Evals

To highlight the reasoning improvement over GPT‑4o, we tested our models on a diverse set of human exams and ML benchmarks. We show that o1 significantly outperforms GPT‑4o on the vast majority of these reasoning-heavy tasks. Unless otherwise specified, we evaluated o1 on the maximal test-time compute setting.

![Competition math eval (AIME 2024)](https://cdn.openai.com/reasoning-evals/v3/headline-mobile-0.png?w=3840&q=90&fm=webp)![Competition math eval (AIME 2024)](https://cdn.openai.com/reasoning-evals/v3/headline-mobile-0-dark.png?w=3840&q=90&fm=webp)

![Competition code (CodeForces)](https://cdn.openai.com/reasoning-evals/v3/headline-mobile-1.png?w=3840&q=90&fm=webp)![Competition code (CodeForces)](https://cdn.openai.com/reasoning-evals/v3/headline-mobile-1-dark.png?w=3840&q=90&fm=webp)

![PhD-Level Science Questio

... (truncated, 35 KB total)