State-of-the-art Small Language Coder Model: Mify-Coder
Abstract
We present Mify-Coder, a 2.5B-parameter code model trained on 4.2T tokens using a compute-optimal strategy built on the Mify-2.5B foundation model. Mify-Coder achieves comparable accuracy and safety while significantly outperforming much larger baseline models on standard coding and function-calling benchmarks, demonstrating that compact models can match frontier-grade models in code generation and agent-driven workflows. Our training pipeline combines high-quality curated sources with synthetic data generated through agentically designed prompts, refined iteratively using enterprise-grade evaluation datasets. LLM-based quality filtering further enhances data density, enabling frugal yet effective training. Through disciplined exploration of CPT-SFT objectives, data mixtures, and sampling dynamics, we deliver frontier-grade code intelligence within a single continuous training trajectory. Empirical evidence shows that principled data and compute discipline allow smaller models to achieve competitive accuracy, efficiency, and safety compliance. Quantized variants of Mify-Coder enable deployment on standard desktop environments without requiring specialized hardware.
Keywords
Cite
@article{arxiv.2512.23747,
title = {State-of-the-art Small Language Coder Model: Mify-Coder},
author = {Abhinav Parmar and Abhisek Panigrahi and Abhishek Kumar Dwivedi and Abhishek Bhattacharya and Adarsh Ramachandra and Aditya Choudhary and Aditya Garg and Aditya Raj and Alankrit Bhatt and Alpesh Yadav and Anant Vishnu and Ananthu Pillai and Ankush Kumar and Aryan Patnaik and Aswatha Narayanan S and Avanish Raj Singh and Bhavya Shree Gadda and Brijesh Pankajbhai Kachhadiya and Buggala Jahnavi and Chidurala Nithin Krishna and Chintan Shah and Chunduru Akshaya and Debarshi Banerjee and Debrup Dey and Deepa R. and Deepika B G and Faiz ur Rahman and Gagan Gayari and Gudhi Jagadeesh Kumar Naidu and Gursimar Singh and Harshal Tyagi and Harshini K and James Mani Vathalloor and Jayarama Nettar and Jayashree Gajjam and Joe Walter Sugil George and Kamalakara Sri Krishna Tadepalli and Kamalkumar Rathinasamy and Karan Chaurasia and Karthikeyan S and Kashish Arora and Kaushal Desai and Khushboo Buwade and Kiran Manjrekar and Malikireddy Venkata Sai Likhitha and Manjunath A and Mitali Mahavir Bedmutha and Mohammed Rafee Tarafdar and Nikhil Tiwari and Nikitha K Gigi and Pavan Ravikumar and Pendyala Swarnanjali and Piyush Anand and Prakash Chandrasekar and Prasanna Bhalchandra Gawade and Prasanth Sivan and Preeti Khurana and Priyanshi Babbar and Rajab Ali Mondal and Rajesh Kumar Vissapragada and Rajeshwari Ganesan and Rajeswari Koppisetti and Ramjee R. and Ramkumar Thiruppathisamy and Rani G. S. and S Reka and Samarth Gupta and Sandeep Reddy Kothakota and Sarathy K and Sathyanarayana Sampath Kumar and Saurabh Kumar and Shashank Khasare and Shenbaga Devi Venkatesh Kumar and Shiva Rama Krishna Parvatham and Shoeb Shaikh and Shrishanmathi A and Shubham Pathak and Sree Samhita Koppaka and Sreenivasa Raghavan K S and Sreeram Venkatasubramanian and Suprabha Desai Bojja and Swetha R and Syed Ahmed and Chinmai Harshitha Thota and Tushar Yadav and Veeravelly Kusumitha and V V S S Prasanth Patnaik and Vidya Sri Sesetti and Vijayakeerthi K and Vikram Raj Bakshi and Vinay K K and Vinoth Kumar Loganathan and Vipin Tiwari and Vivek Kumar Shrivastav and V Venkata Sri Datta Charan and Wasim Akhtar Khan},
journal= {arXiv preprint arXiv:2512.23747},
year = {2026}
}