Justbackup/phi-4

TEXT GENERATIONPricing:Input $0.28 / Output $0.56Concurrent Unit Cost:1Model Size:14.7BQuant:FP8Context Size:32kPublished:Aug 18, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

The phi-4 model by Microsoft Research is a 14.7 billion parameter dense decoder-only Transformer, trained on 9.8 trillion tokens with a 16K token context length. It is built upon a blend of synthetic datasets, filtered public domain websites, and acquired academic books and Q&A datasets, focusing on high-quality data for advanced reasoning. Rigorously enhanced with supervised fine-tuning and direct preference optimization, phi-4 excels in memory/compute constrained environments, latency-bound scenarios, and tasks requiring strong reasoning and logic.

Loading preview...

Overview

Microsoft Research's phi-4 is a 14.7 billion parameter dense decoder-only Transformer model, trained on 9.8 trillion tokens over 21 days using 1920 H100-80G GPUs. It features a 16K token context length and is designed for high-quality reasoning, built upon a unique blend of synthetic, filtered public domain, and academic datasets. The model underwent rigorous enhancement and alignment using supervised fine-tuning (SFT) and direct preference optimization (DPO) to ensure precise instruction adherence and robust safety.

Key Capabilities

  • Advanced Reasoning: Trained with data specifically focused on high quality and advanced reasoning, including math, coding, common sense, and general knowledge.
  • Instruction Adherence & Safety: Incorporates SFT and iterative DPO for precise instruction following and comprehensive safety alignment, evaluated through quantitative benchmarks and qualitative red-teaming.
  • Performance: Demonstrates strong performance across various benchmarks, notably achieving 56.1 on GPQA, 80.4 on MATH, and 82.6 on HumanEval, often outperforming similarly sized models like phi-3 and Qwen 2.5 (14B).
  • Efficiency: Optimized for memory/compute constrained and latency-bound environments.

Good For

  • Accelerating research on language models.
  • Building generative AI features, particularly in English.
  • Applications requiring strong reasoning and logic capabilities.
  • Use cases where memory, compute, or latency are critical factors.

Finetunes
5 models