Pushing boundaries of complex reasoning in small language models

Channel:
Microsoft Research
Subscribers:
351,000
Published on ● Video Link: https://www.youtube.com/watch?v=BCn_Z90pWVM


Duration: 0:00
255 views
19

Mojan Javaheripi, Member of Technical Staff at Microsoft Research AI Frontiers, presents Phi-4-Reasoning and Phi-4-Reasoning-Plus—two 14B models designed to advance complex reasoning in small-scale language models. By introducing a dedicated “thinking block” and applying supervised fine-tuning and reinforcement learning on carefully curated STEM datasets, these models achieve major improvements in problem-solving capabilities.

Phi-4-Reasoning: https://huggingface.co/microsoft/Phi-4-reasoning
Phi-4-Reasoning-Plus: https://huggingface.co/microsoft/Phi-4-reasoning-plus
Phi-4 Reasoning paper (PDF): https://aka.ms/phi4reasoningPDF
Azure AI Foundry Model Catalog: https://aka.ms/AIFoundryModelCatalog
Microsoft/phi-4-gguf on Hugging Face: https://huggingface.co/microsoft/phi-4-gguf

This session aired on September 24, 2025, at Microsoft Research Forum, Season 2 Episode 1.

Register for the series to learn about future episodes: https://aka.ms/registerresearchforumYTs2e1
Continue watching this episode: https://aka.ms/researchforumYTs2e1
Explore all previous episodes: https://aka.ms/researchforumYTplaylist



Other Videos By Microsoft Research


2025-09-24Understanding How Users Prepare for and React to Smartphone Theft
2025-09-24When LLMs Go Online: The Emerging Threat of Web-Enabled LLMs
2025-09-24A Formal Analysis of Apple’s iMessage PQ3 Protocol
2025-09-24Email Spoofing with SMTP Smuggling: How the Shared Email Infrastructures Magnify this Vulnerability
2025-09-24A Framework for Abusability Analysis: The Case of Passkeys in Interpersonal Threat Models
2025-09-24‘Hey mum, I dropped my phone down the toilet’: Investigating Hi Mum and Dad SMS Scams in the UK
2025-09-24Dehumanizing machines: Making sense of AI systems that seem human
2025-09-24Scalable emulation of protein equilibrium ensembles with BioEmu
2025-09-24Disrupting the AI infrastructure with MicroLEDs
2025-09-24Dion: The distributed orthonormal update revolution is here
2025-09-24Pushing boundaries of complex reasoning in small language models
2025-09-22zk-promises: Anonymous Moderation, Reputation, & Blocking from Anonymous Credentials with Callbacks
2025-09-22More is Less: Extra Features in Contactless Payments Break Security
2025-09-18Sub-Population Identification of Multi-morbidity in Sub-Saharan African Populations
2025-09-03Echoes in GenAI generations
2025-08-27Six Years of Rowhammer: Breakthroughs and Future Directions
2025-08-25Sub-Population Identification of Multi-morbidity in Sub-Saharan African Populations
2025-08-19MindJourney: Test-Time Scaling with World Models for Spatial Reasoning
2025-08-11Medical Bayesian Kiosk (2010)
2025-08-07Reimagining healthcare delivery and public health with AI
2025-08-05VeriTrail: Detect hallucination and trace provenance in AI workflows