Phi
Microsoft's small language models that punch above their weight — high-quality reasoning at minimal compute cost.
Overview
Phi is Microsoft Research's family of small language models designed to challenge the assumption that capability requires massive scale. Phi-3 Mini (3.8B parameters) and Phi-3 Small (7B parameters), released in 2024, achieved performance on par with Mistral 7B and Llama 3 8B while being more efficient and suitable for edge deployment.
The key insight behind Phi is data quality over data quantity: rather than training on massive undifferentiated web text, Phi models are trained on carefully curated 'textbook-quality' data — clean, educational, well-reasoned text. This data-centric approach produced models that reason better than models several times larger trained on raw web data.
Phi-3 models run efficiently on smartphones and edge devices, making them important for on-device AI applications where cloud connectivity is unavailable or undesirable. Microsoft has released Phi-3 models on Hugging Face and Azure AI, and they have been integrated into various Azure AI Studio workflows. The success of Phi has influenced the broader field toward examining data quality as a lever distinct from scale.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 8 May 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
ASEAN Perspective
Phi in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Phi is Microsoft's family of small language models that deliver strong reasoning and, in recent versions, multimodal ability at parameter counts that run on laptops, edge and on-device. They are openly available with permissive licensing and are among the best small models for efficiency-per-quality.
They suit developers building local, private or cost-sensitive AI rather than users wanting a chat product; you need ML/inference skills to deploy them. As research/open weights there is no hosted SLA, but they are well documented on Hugging Face and broadly usable worldwide including ASEAN.
Notable facts
- Phi-3 Mini was the first AI model small enough to run entirely on a smartphone to achieve college-level reasoning benchmark performance.
- Microsoft's data curation team identified that training on 'textbook quality' synthetic data generated by GPT-4 dramatically outperformed training on equivalent amounts of raw web text.
- Phi-2 was the first model smaller than 10 billion parameters to outscore models 5x its size on reasoning and commonsense benchmarks.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Phi's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Phi unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Phi directly →
Spotted something out of date? Suggest an update →
Alternatives to Phi
More in LLMs & Chat
