Salvato in:
| Autori principali: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2509.07604 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915493342871552 |
|---|---|
| author | Cheng, Zhoujun Fan, Richard Hao, Shibo Killian, Taylor W. Li, Haonan Sun, Suqi Ren, Hector Moreno, Alexander Zhang, Daqian Zhong, Tianjun Xiong, Yuxin Hu, Yuanzhe Xie, Yutao Han, Xudong Wang, Yuqi Pimpalkhute, Varad Zhuang, Yonghao Singh, Aaryamonvikram Liang, Xuezhi Xie, Anze She, Jianshu Fan, Desai Gao, Chengqian Ma, Liqun Yurochkin, Mikhail Maggs, John Ma, Xuezhe He, Guowei Hu, Zhiting Liu, Zhengzhong Xing, Eric P. |
| author_facet | Cheng, Zhoujun Fan, Richard Hao, Shibo Killian, Taylor W. Li, Haonan Sun, Suqi Ren, Hector Moreno, Alexander Zhang, Daqian Zhong, Tianjun Xiong, Yuxin Hu, Yuanzhe Xie, Yutao Han, Xudong Wang, Yuqi Pimpalkhute, Varad Zhuang, Yonghao Singh, Aaryamonvikram Liang, Xuezhi Xie, Anze She, Jianshu Fan, Desai Gao, Chengqian Ma, Liqun Yurochkin, Mikhail Maggs, John Ma, Xuezhe He, Guowei Hu, Zhiting Liu, Zhengzhong Xing, Eric P. |
| contents | K2-Think is a reasoning system that achieves state-of-the-art performance with a 32B parameter model, matching or surpassing much larger models like GPT-OSS 120B and DeepSeek v3.1. Built on the Qwen2.5 base model, our system shows that smaller models can compete at the highest levels by combining advanced post-training and test-time computation techniques. The approach is based on six key technical pillars: Long Chain-of-thought Supervised Finetuning, Reinforcement Learning with Verifiable Rewards (RLVR), Agentic planning prior to reasoning, Test-time Scaling, Speculative Decoding, and Inference-optimized Hardware, all using publicly available open-source datasets. K2-Think excels in mathematical reasoning, achieving state-of-the-art scores on public benchmarks for open-source models, while also performing strongly in other areas such as Code and Science. Our results confirm that a more parameter-efficient model like K2-Think 32B can compete with state-of-the-art systems through an integrated post-training recipe that includes long chain-of-thought training and strategic inference-time enhancements, making open-source reasoning systems more accessible and affordable. K2-Think is freely available at k2think.ai, offering best-in-class inference speeds of over 2,000 tokens per second per request via the Cerebras Wafer-Scale Engine. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_07604 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | K2-Think: A Parameter-Efficient Reasoning System Cheng, Zhoujun Fan, Richard Hao, Shibo Killian, Taylor W. Li, Haonan Sun, Suqi Ren, Hector Moreno, Alexander Zhang, Daqian Zhong, Tianjun Xiong, Yuxin Hu, Yuanzhe Xie, Yutao Han, Xudong Wang, Yuqi Pimpalkhute, Varad Zhuang, Yonghao Singh, Aaryamonvikram Liang, Xuezhi Xie, Anze She, Jianshu Fan, Desai Gao, Chengqian Ma, Liqun Yurochkin, Mikhail Maggs, John Ma, Xuezhe He, Guowei Hu, Zhiting Liu, Zhengzhong Xing, Eric P. Machine Learning K2-Think is a reasoning system that achieves state-of-the-art performance with a 32B parameter model, matching or surpassing much larger models like GPT-OSS 120B and DeepSeek v3.1. Built on the Qwen2.5 base model, our system shows that smaller models can compete at the highest levels by combining advanced post-training and test-time computation techniques. The approach is based on six key technical pillars: Long Chain-of-thought Supervised Finetuning, Reinforcement Learning with Verifiable Rewards (RLVR), Agentic planning prior to reasoning, Test-time Scaling, Speculative Decoding, and Inference-optimized Hardware, all using publicly available open-source datasets. K2-Think excels in mathematical reasoning, achieving state-of-the-art scores on public benchmarks for open-source models, while also performing strongly in other areas such as Code and Science. Our results confirm that a more parameter-efficient model like K2-Think 32B can compete with state-of-the-art systems through an integrated post-training recipe that includes long chain-of-thought training and strategic inference-time enhancements, making open-source reasoning systems more accessible and affordable. K2-Think is freely available at k2think.ai, offering best-in-class inference speeds of over 2,000 tokens per second per request via the Cerebras Wafer-Scale Engine. |
| title | K2-Think: A Parameter-Efficient Reasoning System |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2509.07604 |