Elamur Research⌕
☰
home / papers / 2025 / Phi-4-reasoning Technical Report

Phi-4-reasoning Technical Report

Paper · 2025 · Marah Abdin, Sahaj Agarwal, Ahmed Awadallah et al.

Summary

The report asks whether a compact model can match much larger systems on hard multi-step reasoning. Microsoft researchers fine-tuned the 14B-parameter Phi-4 on more than 1.4M prompts paired with long reasoning traces written by o3-mini. The result is Phi-4-reasoning. The prompts were chosen to be "teachable": varied in difficulty and close to the limit of what the base model could already handle. They cover STEM, coding and safety, and were decontaminated against the evaluation benchmarks. A second model, Phi-4-reasoning-plus, adds a brief outcome-based RL stage on math problems with verifiable answers. This stage makes its reasoning chains longer.

Both models beat the base model by wide margins and outperform far larger open models such as DeepSeek-R1-Distill-Llama-70B. Phi-4-reasoning-plus comes close to the full DeepSeek-R1 on AIME 2025. Gains also carry over to planning, NP-hard tasks and general benchmarks that training did not target. The work shows that careful data curation plus a light RL step can yield strong small reasoners. It also argues that small benchmarks like AIME need multi-run, variance-aware evaluation.

Key points

  • SFT used over 1.4M curated prompts with o3-mini demonstrations. The RL stage for Phi-4-reasoning-plus used about 6K math problems, and the resulting model uses roughly 1.5× more tokens than Phi-4-reasoning.
  • Compared with Phi-4, the models gain:
    • more than 50 percentage points on AIME 2025 and OmniMath,
    • more than 25 points on LiveCodeBench,
    • 30–60 points on TSP, 3SAT and BA-Calendar.
  • Phi-4-reasoning-plus also improves over Phi-4 on general benchmarks: about +22 on IFEval, +16 on FlenQA and +10 on ArenaHard.
  • On AIME, two runs of average-of-5 evaluation can differ by 5–10 points. The authors therefore report standard deviations and use 50–64 generations per problem on AIME 2025.
  • Known limitations:
    • 32k-token context window,
    • mainly English,
    • SFT data covers only STEM, code and safety, and RL data covers only math,
    • smaller gains in biology, chemistry and discrete math.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,685 works

Cite

Cite as Marah Abdin, Sahaj Agarwal, Ahmed Awadallah, Vidhisha Balachandran, Harkirat Behl, Lingjiao Chen, Gustavo de Rosa, Suriya Gunasekar, Mojan Javaheripi, Neel Joshi, Piero Kauffmann, Yash Lara, Caio César Teodoro Mendes, Arindam Mitra, Besmira Nushi, Dimitris Papailiopoulos, Olli Saarikivi, Shital Shah, Vaishnavi Shrivastava, Vibhav Vineet, Yue Wu, Safoora Yousefi and Guoqing Zheng. 2025. Phi-4-reasoning Technical Report. arXiv:2504.21318.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2504.21318 · arXiv 2504.21318

@misc{abdin2025phi,
  title = {{Phi-4-reasoning Technical Report}},
  author = {Marah Abdin, Sahaj Agarwal, Ahmed Awadallah, Vidhisha Balachandran, Harkirat Behl, Lingjiao Chen, Gustavo {de Rosa}, Suriya Gunasekar, Mojan Javaheripi, Neel Joshi, Piero Kauffmann, Yash Lara, Caio César Teodoro Mendes, Arindam Mitra, Besmira Nushi, Dimitris Papailiopoulos, Olli Saarikivi, Shital Shah, Vaishnavi Shrivastava, Vibhav Vineet, Yue Wu, Safoora Yousefi and Guoqing Zheng},
  year = {2025},
  doi = {10.48550/arxiv.2504.21318},
  eprint = {2504.21318},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2504.21318}
}

References

62 references, 29 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Marah Abdin, Jyoti Aneja, Hany Awadalla, Ahmed Awadallah, Ammar Ahmad Awan, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Jianmin Bao, Harkirat Behl, et al. Phi-3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219, 2024.
  2. Marah Abdin, Jyoti Aneja, Harkirat Behl, Sébastien Bubeck, Ronen Eldan, Suriya Gunasekar, Michael Harrison, Russell J. Hewett, Mojan Javaheripi, Piero Kauffmann, James R. Lee, Yin Tat Lee, Yuanzhi Li, Weishung Liu, Caio C. T. Mendes, Anh Nguyen, Eric Price, Gustavo de Rosa, Olli Saarikivi, Adil Sali… — Phi-4 Technical Report
  3. Marah I Abdin, Suriya Gunasekar, Varun Chandrasekaran, Jerry Li, Mert Yüksekgönül, Rahee Ghosh Peshawaria, Ranjita Naik, and Besmira Nushi. KITAB: evaluating llms on constraint satisfaction for information retrieval. In International Conference on Learning Representations, 2024. find
  4. AIME. Aime 83-24. https://huggingface.co/datasets/lchen001/AIME1983_2024, 2024. Accessed: 2025-03-17. ↗
  5. AIME. Aime 2025. https://huggingface.co/datasets/lchen001/AIME2025, 2025. Accessed: 2025-03-17. ↗
  6. Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, and Dan Mané. Concrete problems in ai safety. arXiv preprint arXiv:1606.06565, 2016.
  7. Anthropic. Claude 3.7 sonnet. https://www.anthropic.com/news/claude-3-7-sonnet, 2025. Accessed: 2025-03-17. ↗
  8. Iván Arcuschin, Jett Janiak, Robert Krzyzanowski, Senthooran Rajamanoharan, Neel Nanda, and Arthur Conmy. Chain-of-thought reasoning in the wild is not always faithful. arXiv preprint arXiv:2503.08679, 2025. find
  9. Vidhisha Balachandran, Jingya Chen, Neel Joshi, Besmira Nushi, Hamid Palangi, Eduardo Salinas, Vibhav Vineet, James Woffinden-Luey, and Safoora Yousefi. Eureka: Evaluating and understanding large foundation models. arXiv preprint arXiv:2409.10566, 2024. find
  10. Vidhisha Balachandran, Jingya Chen, Lingjiao Chen, Shivam Garg, Neel Joshi, Yash Lara, John Langford, Besmira Nushi, Vibhav Vineet, Yue Wu, and Safoora Yousefi. Inference-time scaling for complex tasks: Where we stand and what lies ahead, 2025. URL https://arxiv.org/abs/2504.00294. ↗
  11. Mislav Balunović, Jasper Dekoninck, Ivo Petrov, Nikola Jovanović, and Martin Vechev. Matharena: Evaluating llms on uncontaminated math competitions, February 2025. URL https://matharena.ai/. ↗
  12. Solon Barocas, Anhong Guo, Ece Kamar, Jacquelyn Krones, Meredith Ringel Morris, Jennifer Wortman Vaughan, W Duncan Wadsworth, and Hanna Wallach. Designing disaggregated evaluations of ai systems: Choices, considerations, and tradeoffs. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, an… find
  13. Natasha Butt, Varun Chandrasekaran, Neel Joshi, Besmira Nushi, and Vidhisha Balachandran. Benchagents: Automated benchmark creation with agent interaction. arXiv preprint arXiv:2410.22584, 2024. — [arXiv:2410.22584](https://arxiv.org/abs/2410.22584) find
  14. Shouyuan Chen, Sherman Wong, Liangjian Chen, and Yuandong Tian. Extending context window of large language models via positional interpolation. arXiv preprint arXiv:2306.15595, 2023. find
  15. Quy-Anh Dang and Chris Ngo. Reinforcement learning for reasoning in small llms: What works and what doesn’t, 2025. URL https://arxiv.org/abs/2503.16219. ↗
  16. Bofei Gao, Feifan Song, Zhe Yang, Zefan Cai, Yibo Miao, Qingxiu Dong, Lei Li, Chenghao Ma, Liang Chen, Runxin Xu, et al. Omni-math: A universal olympiad level mathematic benchmark for large language models. ICLR, 2025.
  17. Leo Gao, John Schulman, and Jacob Hilton. Scaling laws for reward model overoptimization. In International Conference on Machine Learning, pages 10835–10866. PMLR, 2023.
  18. Google. Gemini flash thinking. https://deepmind.google/technologies/gemini/flash/, 2025. Accessed: 2025-03-17. ↗
  19. Xinyu Guan, Li Lyna Zhang, Yifei Liu, Ning Shang, Youran Sun, Yi Zhu, Fan Yang, and Mao Yang. rstar-math: Small llms can master math reasoning with self-evolved deep thinking. arXiv preprint arXiv:2501.04519, 2025.
  20. Suriya Gunasekar, Yi Zhang, Jyoti Aneja, Caio César Teodoro Mendes, Allie Del Giorno, Sivakanth Gopi, Mojan Javaheripi, Piero Kauffmann, Gustavo de Rosa, Olli Saarikivi, et al. Textbooks are all you need. arXiv preprint arXiv:2306.11644, 2023. find
  21. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948, 2025.
  22. Juris Hartmanis. Computers and intractability: a guide to the theory of np-completeness (michael r. garey and david s. johnson). Siam Review, 24(1):90, 1982. find
  23. Thomas Hartvigsen, Saadia Gabriel, Hamid Palangi, Maarten Sap, Dipankar Ray, and Ece Kamar. ToxiGen: A large-scale machine-generated dataset for adversarial and implicit hate speech detection. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long…
  24. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding, 2021. URL https://arxiv.org/abs/2009.03300. ↗
  25. Andreas Hochlehnert, Hardik Bhatnagar, Vishaal Udandarao, Samuel Albanie, Ameya Prabhu, and Matthias Bethge. A sober look at progress in language model reasoning: Pitfalls and paths to reproducibility. arXiv preprint arXiv:2504.07086, 2025. find
  26. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card. arXiv preprint arXiv:2410.21276, 2024.
  27. Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, et al. Openai o1 system card. arXiv preprint arXiv:2412.16720, 2024.
  28. Mojan Javaheripi, Sébastien Bubeck, Marah Abdin, Jyoti Aneja, Caio César Teodoro Mendes, Weizhu Chen, Allie Del Giorno, Ronen Eldan, Sivakanth Gopi, Suriya Gunasekar, Piero Kauffmann, Yin Tat Lee, Yuanzhi Li, Anh Nguyen, Gustavo de Rosa, Olli Saarikivi, Adil Salim, Shital Shah, Michael Santacroce, H… find
  29. Fengqing Jiang, Zhangchen Xu, Yuetai Li, Luyao Niu, Zhen Xiang, Bo Li, Bill Yuchen Lin, and Radha Poovendran. Safechain: Safety of language models with long chain-of-thought reasoning capabilities. arXiv preprint arXiv:2502.12025, 2025. find
  30. Mosh Levy, Alon Jacoby, and Yoav Goldberg. Same task, more tokens: the impact of input length on the reasoning performance of large language models. In ACL, 2024.
  31. LG AI Research. Exaone deep: Reasoning enhanced language models. arXiv preprint arXiv:2503.12524, 2025. — [arXiv:2503.12524](https://arxiv.org/abs/2503.12524) find
  32. Shanda Li, Chong You, Guru Guruganesh, Joshua Ainslie, Santiago Ontanon, Manzil Zaheer, Sumit Sanghai, Yiming Yang, Sanjiv Kumar, and Srinadh Bhojanapalli. Functional interpolation for relative positions improves long context transformers. arXiv preprint arXiv:2310.04418, 2023. find
  33. Tianle Li, Wei-Lin Chiang, Evan Frick, Lisa Dunlap, Tianhao Wu, Banghua Zhu, Joseph E Gonzalez, and Ion Stoica. From crowdsourced data to high-quality benchmarks: Arena-hard and benchbuilder pipeline. arXiv preprint arXiv:2406.11939, 2024.
  34. Xuefeng Li, Haoyang Zou, and Pengfei Liu. Limr: Less is more for rl scaling, 2025. find
  35. Jiawei Liu, Chunqiu Steven Xia, Yuyao Wang, and Lingming Zhang. Is your code generated by chatgpt really correct? rigorous evaluation of large language models for code generation, 2023. URL https://arxiv.org/abs/2305. 01210. ↗
  36. Michael Luo, Sijun Tan, Justin Wong, Xiaoxiang Shi, William Y. Tang, Manan Roongta, Colin Cai, Jeffrey Luo, Li Erran Li, Raluca Ada Popa, and Ion Stoica. Deepscaler: Surpassing o1-preview with a 1.5b model by scaling rl. https://pretty-radio-b75.notion.site/ DeepScaleR-Surpassing-O1-Preview-with-a-1… ↗
  37. Ahmed Magooda, Alec Helyar, Kyle Jackson, David Sullivan, Chad Atalla, Emily Sheng, Dan Vann, Richard Edgar, Hamid Palangi, Roman Lutz, Hongliang Kong, Vincent Yun, Eslam Kamal, Federico Zarfati, Hanna Wallach, Sarah Bird, and Mei Chen. A framework for automated measurement of responsible ai harms i… — A Framework for Automated Measurement of Responsible AI Harms in Generative AI Applications
  38. Arindam Mitra, Luciano Del Corro, Shweti Mahajan, Andres Codas, Clarisse Simoes, Sahaj Agarwal, Xuxi Chen, Anastasia Razdaibiedina, Erik Jones, Kriti Aggarwal, et al. Orca 2: Teaching small language models how to reason. arXiv preprint arXiv:2311.11045, 2023.
  39. Arindam Mitra, Luciano Del Corro, Guoqing Zheng, Shweti Mahajan, Dany Rouhana, Andres Codas, Yadong Lu, Wei-ge Chen, Olga Vrousgos, Corby Rosset, et al. Agentinstruct: Toward generative teaching with agentic flows. arXiv preprint arXiv:2407.03502, 2024. find
  40. Mazda Moayeri, Vidhisha Balachandran, Varun Chandrasekaran, Safoora Yousefi, Thomas Fel, Soheil Feizi, Besmira Nushi, Neel Joshi, and Vibhav Vineet. Unearthing skill-level insights for understanding trade-offs of foundation models. ICLR, 2025. find
  41. Subhabrata Mukherjee, Arindam Mitra, Ganesh Jawahar, Sahaj Agarwal, Hamid Palangi, and Ahmed Awadallah. Orca: Progressive learning from complex explanation traces of gpt-4. arXiv preprint arXiv:2306.02707, 2023.
  42. Besmira Nushi, Ece Kamar, and Eric Horvitz. Towards accountable ai: Hybrid human-machine analyses for characterizing system failure. In Proceedings of the AAAI Conference on Human Computation and Crowdsourcing, volume 6, pages 126–135, 2018. find
  43. OpenAI. Openai o3-mini system card. https://openai.com/index/o3-mini-system-card/, 2025. Accessed: 2025-03-17. ↗
  44. Christos H Papadimitriou. Computational complexity. In Encyclopedia of computer science, pages 260–265. John Wiley and Sons Ltd., 2003. find
  45. Samir Passi and Mihaela Vorvoreanu. Overreliance on ai literature review. Microsoft Research, 339:340, 2022. find
  46. Ivo Petrov, Jasper Dekoninck, Lyuben Baltadzhiev, Maria Drencheva, Kristian Minchev, Mislav Balunović, Nikola Jovanović, and Martin Vechev. Proof or bluff? evaluating llms on 2025 usa math olympiad. arXiv preprint arXiv:2503.21934, 2025. find
  47. David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R Bowman. Gpqa: A graduate-level google-proof q&a benchmark. In First Conference on Language Modeling, 2024.
  48. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300, 2024.
  49. Guangming Sheng, Chi Zhang, Zilingfeng Ye, Xibin Wu, Wang Zhang, Ru Zhang, Yanghua Peng, Haibin Lin, and Chuan Wu. Hybridflow: A flexible and efficient rlhf framework. arXiv preprint arXiv: 2409.19256, 2024.
  50. Freda Shi, Mirac Suzgun, Markus Freitag, Xuezhi Wang, Suraj Srivats, Soroush Vosoughi, Hyung Won Chung, Yi Tay, Sebastian Ruder, Denny Zhou, Dipanjan Das, and Jason Wei. Language models are multilingual chain-of-thought reasoners, 2022. URL https://arxiv.org/abs/2210.03057. ↗
  51. Jianlin Su, Yu Lu, Shengfeng Pan, Ahmed Murtadha, Bo Wen, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding. arXiv preprint arXiv:2104.09864, 2021.
  52. Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, et al. Kimi k1. 5: Scaling reinforcement learning with llms. arXiv preprint arXiv:2501.12599, 2025.
  53. OpenThoughts Team. Open Thoughts. https://open-thoughts.ai, January 2025. ↗
  54. Qwen Team. Qwq-32b: Embracing the power of reinforcement learning, March 2025. URL https://qwenlm.github.io/blog/qwq-32b/. ↗
  55. Miles Turpin, Julian Michael, Ethan Perez, and Samuel R. Bowman. Language models don’t always say what they think: Unfaithful explanations in chain-of-thought prompting, 2023. URL https://arxiv.org/abs/2305.04388. ↗
  56. Jiayu Wang, Yifei Ming, Zhenmei Shi, Vibhav Vineet, Xin Wang, Sharon Li, and Neel Joshi. Is a picture worth a thousand words? delving into spatial reasoning for vision language models. Advances in Neural Information Processing Systems, 37:75392–75421, 2024. find
  57. Yubo Wang, Xueguang Ma, Ge Zhang, Yuansheng Ni, Abhranil Chandra, Shiguang Guo, Weiming Ren, Aaran Arulraj, Xuan He, Ziyan Jiang, Tianle Li, Max Ku, Kai Wang, Alex Zhuang, Rongqi Fan, Xiang Yue, and Wenhu Chen. Mmlu-pro: A more robust and challenging multi-task language understanding benchmark, 2024…
  58. An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, et al. Qwen2. 5 technical report. arXiv preprint arXiv:2412.15115, 2024. find
  59. Guanghao Ye, Khiem Duc Pham, Xinzhi Zhang, Sivakanth Gopi, Baolin Peng, Beibin Li, Janardhan Kulkarni, and Huseyin A Inan. On the emergence of thinking in llms i: Searching for the right intuition. arXiv preprint arXiv:2502.06773, 2025. — [arXiv:2502.06773](https://arxiv.org/abs/2502.06773) find
  60. Edward Yeo, Yuxuan Tong, Morry Niu, Graham Neubig, and Xiang Yue. Demystifying long chain-of-thought reasoning in llms. arXiv preprint arXiv:2502.03373, 2025.
  61. Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Tiantian Fan, Gaohong Liu, Lingjun Liu, Xin Liu, Haibin Lin, Zhiqi Lin, Bole Ma, Guangming Sheng, Yuxuan Tong, Chi Zhang, Mofan Zhang, Wang Zhang, Hang Zhu, Jinhua Zhu, Jiaze Chen, Jiangjie Chen, Chengyi Wang, Hongli Yu, Weinan D… — DAPO: An Open-Source LLM Reinforcement Learning System at Scale
  62. Jeffrey Zhou, Tianjian Lu, Swaroop Mishra, Siddhartha Brahma, Sujoy Basu, Yi Luan, Denny Zhou, and Le Hou. Instruction-following evaluation for large language models. arXiv preprint arXiv:2311.07911, 2023.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/abdin-2025-phi-4-reasoning-technical-report/