Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2502.06807 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916620021006336 |
|---|---|
| author | OpenAI : El-Kishky, Ahmed Wei, Alexander Saraiva, Andre Minaiev, Borys Selsam, Daniel Dohan, David Song, Francis Lightman, Hunter Clavera, Ignasi Pachocki, Jakub Tworek, Jerry Kuhn, Lorenz Kaiser, Lukasz Chen, Mark Schwarzer, Max Rohaninejad, Mostafa McAleese, Nat contributors, o3 Mürk, Oleg Garg, Rhythm Shu, Rui Sidor, Szymon Kosaraju, Vineet Zhou, Wenda |
| author_facet | OpenAI : El-Kishky, Ahmed Wei, Alexander Saraiva, Andre Minaiev, Borys Selsam, Daniel Dohan, David Song, Francis Lightman, Hunter Clavera, Ignasi Pachocki, Jakub Tworek, Jerry Kuhn, Lorenz Kaiser, Lukasz Chen, Mark Schwarzer, Max Rohaninejad, Mostafa McAleese, Nat contributors, o3 Mürk, Oleg Garg, Rhythm Shu, Rui Sidor, Szymon Kosaraju, Vineet Zhou, Wenda |
| contents | We show that reinforcement learning applied to large language models (LLMs) significantly boosts performance on complex coding and reasoning tasks. Additionally, we compare two general-purpose reasoning models - OpenAI o1 and an early checkpoint of o3 - with a domain-specific system, o1-ioi, which uses hand-engineered inference strategies designed for competing in the 2024 International Olympiad in Informatics (IOI). We competed live at IOI 2024 with o1-ioi and, using hand-crafted test-time strategies, placed in the 49th percentile. Under relaxed competition constraints, o1-ioi achieved a gold medal. However, when evaluating later models such as o3, we find that o3 achieves gold without hand-crafted domain-specific strategies or relaxed constraints. Our findings show that although specialized pipelines such as o1-ioi yield solid improvements, the scaled-up, general-purpose o3 model surpasses those results without relying on hand-crafted inference heuristics. Notably, o3 achieves a gold medal at the 2024 IOI and obtains a Codeforces rating on par with elite human competitors. Overall, these results indicate that scaling general-purpose reinforcement learning, rather than relying on domain-specific techniques, offers a robust path toward state-of-the-art AI in reasoning domains, such as competitive programming. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_06807 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Competitive Programming with Large Reasoning Models OpenAI : El-Kishky, Ahmed Wei, Alexander Saraiva, Andre Minaiev, Borys Selsam, Daniel Dohan, David Song, Francis Lightman, Hunter Clavera, Ignasi Pachocki, Jakub Tworek, Jerry Kuhn, Lorenz Kaiser, Lukasz Chen, Mark Schwarzer, Max Rohaninejad, Mostafa McAleese, Nat contributors, o3 Mürk, Oleg Garg, Rhythm Shu, Rui Sidor, Szymon Kosaraju, Vineet Zhou, Wenda Machine Learning Artificial Intelligence Computation and Language We show that reinforcement learning applied to large language models (LLMs) significantly boosts performance on complex coding and reasoning tasks. Additionally, we compare two general-purpose reasoning models - OpenAI o1 and an early checkpoint of o3 - with a domain-specific system, o1-ioi, which uses hand-engineered inference strategies designed for competing in the 2024 International Olympiad in Informatics (IOI). We competed live at IOI 2024 with o1-ioi and, using hand-crafted test-time strategies, placed in the 49th percentile. Under relaxed competition constraints, o1-ioi achieved a gold medal. However, when evaluating later models such as o3, we find that o3 achieves gold without hand-crafted domain-specific strategies or relaxed constraints. Our findings show that although specialized pipelines such as o1-ioi yield solid improvements, the scaled-up, general-purpose o3 model surpasses those results without relying on hand-crafted inference heuristics. Notably, o3 achieves a gold medal at the 2024 IOI and obtains a Codeforces rating on par with elite human competitors. Overall, these results indicate that scaling general-purpose reinforcement learning, rather than relying on domain-specific techniques, offers a robust path toward state-of-the-art AI in reasoning domains, such as competitive programming. |
| title | Competitive Programming with Large Reasoning Models |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2502.06807 |