Microsoft Small Language Models — Phi & Aion
⚡ Microsoft Small Language Models — Phi & Aion
Section titled “⚡ Microsoft Small Language Models — Phi & Aion”Microsoft pioneered the idea that a small model trained on the right data can outperform models many times its size. The Phi family proved this at scale. At Microsoft Build 2026, Microsoft went further — shipping Aion 1.0 directly into Windows as an on-device model that runs on CPU alone, with no cloud required.
This page covers every Microsoft SLM from the original Phi-1 in 2023 through the Aion 1.0 models in 2026.
📊 Complete Microsoft SLM Overview
Section titled “📊 Complete Microsoft SLM Overview”| 🤖 Model | 📅 Released | ⚙️ Parameters | 📄 Context | 🎯 Focus |
|---|---|---|---|---|
| ⚡ Phi-1 | Jun 2023 | 1.3B | 2K | Python coding |
| ⚡ Phi-1.5 | Sep 2023 | 1.3B | 2K | Common sense reasoning |
| ⚡ Phi-2 | Dec 2023 | 2.7B | 2K | Language + reasoning |
| ⚡ Phi-3-mini | Apr 2024 | 3.8B | 4K / 128K | General language |
| ⚡ Phi-3-small | May 2024 | 7B | 8K / 128K | General language |
| ⚡ Phi-3-medium | May 2024 | 14B | 4K / 128K | General language |
| 👁️ Phi-3-vision | May 2024 | 4.2B | 128K | Language + vision |
| ⚡ Phi-3.5-mini | Aug 2024 | 3.82B | 128K | General language |
| 🧠 Phi-3.5-MoE | Aug 2024 | 41.9B (MoE) | 128K | Advanced reasoning |
| 👁️ Phi-3.5-vision | Aug 2024 | 4.15B | 128K | Image + video analysis |
| 🧠 Phi-4 | Dec 2024 | 14B | 16K | Complex reasoning + math |
| ⚡ Phi-4-mini | Feb 2025 | 3.8B | 128K | Speed + long context |
| 🎭 Phi-4-multimodal | Feb 2025 | 5.6B | 128K | Speech + vision + text |
| 🔬 Phi-4-reasoning | Apr 2025 | 14B | 128K | Deep reasoning (fine-tuned) |
| 🪟 Aion 1.0 Instruct | Jun 2026 | Undisclosed | — | On-device Windows tasks |
| 🪟 Aion 1.0 Plan | Jun 2026 | 14B | 32K | On-device reasoning + agents |
🔢 Phi-1 and Phi-2 — Where It All Began (2023)
Section titled “🔢 Phi-1 and Phi-2 — Where It All Began (2023)”⚡ Phi-1 — June 2023
Phi-1 was Microsoft’s proof-of-concept that model quality is more about data quality than raw model size.
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | June 2023 |
| ⚙️ Parameters | 1.3 billion |
| 📄 Context window | 2,048 tokens |
| 🎯 Focus | Python code generation |
| 📜 License | MIT (open weights) |
✅ What it showed: A 1.3B model trained on carefully curated “textbook-quality” Python data could match or beat much larger code models on Python benchmarks.
🎯 Primary use: Python code generation, code understanding, programming exercises.
⚡ Phi-1.5 — September 2023
Phi-1.5 extended the Phi-1 experiment to general language and common sense reasoning.
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | September 2023 |
| ⚙️ Parameters | 1.3 billion |
| 📄 Context window | 2,048 tokens |
| 🎯 Focus | Common sense reasoning, language |
| 📜 License | MIT (open weights) |
✅ What it showed: The “textbook data” approach generalised beyond code to broader reasoning tasks.
⚡ Phi-2 — December 2023
Phi-2 doubled the parameter count and demonstrated that a 2.7B model could punch well above its weight class.
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | December 2023 |
| ⚙️ Parameters | 2.7 billion |
| 📄 Context window | 2,048 tokens |
| 🎯 Focus | Language understanding + reasoning |
| 📜 License | MIT (open weights) |
✅ What it showed: Phi-2 outperformed Llama-2-7B and Mistral-7B on several benchmarks — a 2.7B model beating 7B models.
🎯 Primary use: Lightweight assistants, classification, summarisation, local deployment experiments.
🔢 Phi-3 Family — The Production SLM (2024)
Section titled “🔢 Phi-3 Family — The Production SLM (2024)”The Phi-3 family was Microsoft’s first SLM family designed for real production deployment — not just research. It runs on phones, laptops, and edge devices.
⚡ Phi-3-mini — April 2024
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | April 2024 |
| ⚙️ Parameters | 3.8 billion |
| 📄 Context window | 4K tokens (standard) · 128K tokens (long) |
| 🎯 Focus | General language, reasoning |
| 🌐 Availability | Azure AI · Hugging Face · ONNX · NVIDIA NIM |
| 📜 License | MIT (open weights) |
✅ Best for:
- 📱 Mobile and edge device deployment
- 🏷️ Classification and routing
- 📝 Short-form summarisation
- ⚡ Low-latency assistants
💡 Key result: Phi-3-mini outperforms models twice its size on several language and reasoning benchmarks.
⚡ Phi-3-small — May 2024
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | May 2024 |
| ⚙️ Parameters | 7 billion |
| 📄 Context window | 8K tokens (standard) · 128K tokens (long) |
| 🎯 Focus | General language, stronger reasoning |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
✅ Best for:
- 🧠 Tasks needing more reasoning than Phi-3-mini can handle
- 📄 Longer context processing
- 💬 Production chat assistants at moderate cost
💡 Key result: Phi-3-small outperforms GPT-3.5T on many benchmarks despite being a fraction of the size.
⚡ Phi-3-medium — May 2024
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | May 2024 |
| ⚙️ Parameters | 14 billion |
| 📄 Context window | 4K tokens (standard) · 128K tokens (long) |
| 🎯 Focus | Strong general language and reasoning |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
✅ Best for:
- 🧠 Tasks requiring stronger reasoning than 7B models
- 💻 Code understanding and explanation
- 📄 Document analysis
- 🏢 Enterprise RAG and assistants
👁️ Phi-3-vision — May 2024
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | May 2024 |
| ⚙️ Parameters | 4.2 billion |
| 📄 Context window | 128K tokens |
| 🎯 Focus | Language + image understanding |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
✅ Best for:
- 👁️ Image analysis and visual Q&A
- 🖥️ Screenshot understanding
- 📊 Chart and diagram explanation
- 🔤 OCR-style extraction from images
🔢 Phi-3.5 Family — Extended Capabilities (August 2024)
Section titled “🔢 Phi-3.5 Family — Extended Capabilities (August 2024)”The Phi-3.5 series brought three new models, notably the MoE variant and improved vision capabilities.
⚡ Phi-3.5-mini — August 2024
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | August 2024 |
| ⚙️ Parameters | 3.82 billion |
| 📄 Context window | 128K tokens |
| 🎯 Focus | Fast general language, long context |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
✅ Best for:
- 📄 Long document summarisation
- 🔍 Information retrieval over long contexts
- ⚡ Low-latency production assistants
- 📱 Mobile and edge deployment
🧠 Phi-3.5-MoE — August 2024
The Mixture-of-Experts variant — Microsoft’s most capable Phi-3.5 model.
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | August 2024 |
| ⚙️ Parameters | 41.9 billion (MoE — not all active at once) |
| 📄 Context window | 128K tokens |
| 🎯 Focus | Advanced reasoning, strong general tasks |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
💡 What MoE means: Only a subset of the 41.9B parameters are active for any given token — giving large-model quality at lower inference cost.
✅ Best for:
- 🧠 Complex multi-step reasoning
- 💻 Advanced code tasks
- 📄 Long document analysis
- 🧩 Tasks requiring both breadth and depth
👁️ Phi-3.5-vision — August 2024
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | August 2024 |
| ⚙️ Parameters | 4.15 billion |
| 📄 Context window | 128K tokens |
| 🎯 Focus | Image + video analysis |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
✅ Best for:
- 🖼️ Image and video content understanding
- 📊 Visual data extraction
- 🖥️ UI and screenshot analysis
- 🎬 Video frame-level Q&A
🔢 Phi-4 Family — Reasoning-First SLMs (December 2024 – April 2025)
Section titled “🔢 Phi-4 Family — Reasoning-First SLMs (December 2024 – April 2025)”Phi-4 represented a major leap: reasoning ability became a first-class design goal, not just benchmark performance.
🧠 Phi-4 — December 2024
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | December 2024 |
| ⚙️ Parameters | 14 billion |
| 📄 Context window | 16,384 tokens |
| 🎯 Focus | Complex reasoning · mathematics · STEM |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
✅ What makes it special:
- 🔢 State-of-the-art on math benchmarks at 14B scale
- 🧠 Significantly better at multi-step reasoning than Phi-3-medium
- 💻 Strong coding and code explanation ability
💡 Example:
“Prove that the sum of two odd numbers is always even, and generalise the proof to show why the product of two odd numbers is always odd.”
⚡ Phi-4-mini — February 2025
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | February 26, 2025 |
| ⚙️ Parameters | 3.8 billion |
| 📄 Context window | 128,000 tokens |
| 🎯 Focus | Speed · long context · efficiency |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
| 🔤 Vocabulary | 200,000 tokens (grouped-query attention) |
✅ Best for:
- 📄 Long document summarisation (up to 128K tokens)
- ⚡ High-speed classification and extraction
- 📱 Edge and mobile deployment
- 🔍 Information retrieval and RAG assistants
💡 Key fact: At 3.8B parameters with a 128K context window, Phi-4-mini handles document lengths that many larger models cannot.
🎭 Phi-4-multimodal — February 2025
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | February 26, 2025 |
| ⚙️ Parameters | 5.6 billion |
| 📄 Context window | 128,000 tokens |
| 🎯 Focus | Speech + vision + text — all three at once |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
✅ What makes it special:
- 🎙️ Processes speech audio directly (no separate transcription step)
- 👁️ Understands images and vision inputs
- 💬 Handles text — all three modalities in one model
💡 Example use case:
A field technician speaks a question while pointing their phone camera at a broken circuit board → Phi-4-multimodal transcribes the speech, analyses the image, and returns a repair instruction — all on-device.
🔬 Phi-4-reasoning — April 2025
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Released | April 30, 2025 |
| ⚙️ Parameters | 14 billion |
| 📄 Context window | 128,000 tokens |
| 🎯 Focus | Deep reasoning — fine-tuned from Phi-4 |
| 🌐 Availability | Azure AI · Hugging Face |
| 📜 License | MIT (open weights) |
✅ What makes it special:
- 🔬 Fine-tuned specifically to improve chain-of-thought reasoning
- 🔢 Excellent on maths, logic, and scientific problem solving
- 🧠 Best Microsoft SLM for tasks where you would otherwise reach for a large reasoning model
💡 Example:
“A train leaves Station A at 60 km/h. Another leaves Station B (300 km away) at 90 km/h toward the same destination. A bird flies at 200 km/h back and forth between the trains until they meet. How far does the bird fly?”
🪟 Aion 1.0 — On-Device AI Built Into Windows (June 2026)
Section titled “🪟 Aion 1.0 — On-Device AI Built Into Windows (June 2026)”Announced at Microsoft Build 2026, Aion 1.0 is a new generation of on-device SLMs that ship inside Windows 11. Unlike cloud models, Aion runs locally — no internet connection required, no API call, no data leaving the device.
🪟 Aion 1.0 Instruct — The Everyday Workhorse
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Announced | June 2, 2026 (Microsoft Build 2026) |
| ⚙️ Parameters | Not published (described as “smaller, faster, more efficient” than the previous Windows OS SLM) |
| 🏃 Hardware | Runs on CPU — no NPU or GPU required |
| 🚀 Open source | Planned for Hugging Face release — July 2026 |
| 🌐 Preview | Available now in Microsoft Edge Insider channels |
✅ What it does:
- 📝 Summarisation of documents and web pages
- ✏️ Text rewriting and tone adjustment
- 🏷️ Intent detection
- ♿ Accessibility features (in-OS)
- 🖥️ Works on almost any Windows PC — no special hardware needed
💡 What makes it different:
Because it runs entirely on CPU, Aion 1.0 Instruct works on far more Windows devices than any previous on-device AI model — including older PCs without an NPU or dedicated GPU.
🧠 Aion 1.0 Plan — On-Device Reasoning and Agents
| 🔧 Property | 📋 Detail |
|---|---|
| 📅 Announced | June 2, 2026 (Microsoft Build 2026) |
| ⚙️ Parameters | 14 billion |
| 📄 Context window | 32,000 tokens |
| 🎯 Focus | Reasoning · tool calling · agent orchestration |
| 🏃 Hardware | Runs on-device (Windows 11 built-in) |
| 🌐 Access | Ships in-box as part of Windows |
✅ What it does:
- 🧠 Reasons over user intent without sending data to the cloud
- 🔧 Invokes local tools (file management, app control, OS functions)
- 📁 Manages files and system tasks via agent loops
- 🤖 Orchestrates sub-agents on the local device
- 🔒 Fully private — no data leaves the machine
💡 Example use case:
“Organise all PDFs in my Downloads folder from the last 30 days by project name, create a subfolder for each project, and generate a markdown summary of each file.”
Aion 1.0 Plan handles this entirely on-device — no cloud, no API, no data leaving the PC.
🎯 Choosing the Right Microsoft SLM
Section titled “🎯 Choosing the Right Microsoft SLM”| 🎯 Task | ✅ Best Model | 💡 Why |
|---|---|---|
| ⚡ High-speed classification or routing | Phi-3-mini or Phi-4-mini | Smallest models with excellent classification ability |
| 📄 Long document summarisation (up to 128K) | Phi-4-mini or Phi-3.5-mini | Large context window at small model size |
| 🔢 Maths, logic, or STEM reasoning | Phi-4-reasoning | Fine-tuned specifically for deep reasoning |
| 🏗️ Complex reasoning at low cost | Phi-3.5-MoE | MoE architecture gives large-model quality |
| 👁️ Image or screenshot analysis | Phi-3.5-vision or Phi-4-multimodal | Built-in vision understanding |
| 🎙️ Speech + image + text in one model | Phi-4-multimodal | Only Microsoft SLM handling all three |
| 📱 Mobile or edge deployment | Phi-3-mini, Phi-4-mini | Smallest and most deployment-friendly |
| 🏠 Fully offline, no cloud, on Windows | Aion 1.0 Instruct | Runs on CPU, built into Windows |
| 🤖 On-device agent and tool calling | Aion 1.0 Plan | 14B reasoning model built into Windows |
| 💻 Code generation (general) | Phi-4 or Phi-3-medium | Strong coding + reasoning combination |
🌐 Where to Access Microsoft SLMs
Section titled “🌐 Where to Access Microsoft SLMs”| 🏷️ Platform | 🤖 Models Available | 🔗 Notes |
|---|---|---|
| 🔷 Azure AI Foundry | Full Phi family | API and managed deployment |
| 🤗 Hugging Face | Full Phi family | Free download, MIT licence |
| 🏠 ONNX Runtime | Phi-3 family | Optimised for Windows, mobile, web |
| 🟩 NVIDIA NIM | Phi-3 family | Inference on NVIDIA GPUs |
| 🔵 Intel accelerators | Phi-3 family | Optimised inference |
| 🪟 Windows 11 built-in | Aion 1.0 Instruct, Aion 1.0 Plan | Ships in-box, no API key required |
| 🌐 Microsoft Edge Insider | Aion 1.0 Instruct | Preview available now |
💡 Why Microsoft SLMs Are Different
Section titled “💡 Why Microsoft SLMs Are Different”| ✅ Advantage | 📋 Explanation |
|---|---|
| 📜 Open weights (MIT) | All Phi models are freely downloadable and modifiable |
| 🏠 Local deployment | Designed to run on phones, laptops, and edge devices |
| 🔒 Privacy | On-device models (Aion) keep data on the machine |
| 💰 Cost efficiency | Smaller models = lower inference cost at scale |
| 🎓 “Textbook data” training | Quality of training data over quantity of raw text |
| 🌐 Broad hardware support | ONNX, NVIDIA NIM, Intel — not just Azure GPUs |