Skip to content
AI Model Radar

Phi 4 mini instruct

microsoft/Phi-4-mini-instruct
Open Source · chat · open-weights
GA Alert me on changes
Context
131.1K
Max output
128K
Input $/1M
$0.08
Output $/1M
$0.35
Modalities
text
Released
19 Feb 2025
License: mit · microsoft/Phi-4-mini-instruct
AI summary
● machine-written

Microsoft releases Phi-4 Mini Instruct, lightweight instructed model

Phi-4-mini-instruct is a lightweight open-weights model built on synthetic data and filtered public websites, focused on reasoning-dense content. The model supports a 131K token context window and has undergone supervised fine-tuning and direct preference optimization for instruction adherence and safety. It is available under the MIT license and suitable for scenarios requiring compact models with strong reasoning capabilities.

What's new
  • Context window of 131,072 tokens (131K)
  • Trained on synthetic data and filtered publicly available websites
  • Incorporates supervised fine-tuning and direct preference optimization
  • Open-weights model available under MIT license
Best for
High-quality reasoning tasksInstruction-following applicationsDeployments with limited memory requirementsScenarios requiring compact model footprint
Sources

Source: https://huggingface.co/microsoft/Phi-4-mini-instruct