Phi-3
Microsoft's Small Models, Big Quality
Microsoft's series of small foundation models — 3.8B / 7B / 14B parameters — that punched far above their weight by training on aggressively filtered "textbook-quality" data plus synthetic examples generated by GPT-4. The 3.8B Mini variant reaches GPT-3.5-class quality while running on a smartphone — making it the foundation of much on-device AI in 2024-25.
Cost
Free
Open weights — self-host
Context
128K
Up to 128,000 tokens
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on
Artificial Analysis. Cost buckets are fixed dollar
thresholds keyed off output-token price ($/M out).
Intelligence
- Top 1%≤ 1%
- Top 5%≤ 5%
- Top 10%≤ 10%
- Good≤ 25%
- Medium≤ 50%
- Below avg> 50%
Speed
- Top 1%≥ 345 tok/s
- Top 5%≥ 237 tok/s
- Top 10%≥ 196 tok/s
- Good≥ 146 tok/s
- Medium≥ 90 tok/s
- Slow< 90 tok/s
Cost
- Freeopen weights · self-host
- Low< $1 / M out
- Moderate$1–5 / M out
- High≥ $5 / M out
Why it matters
Phi-3 is the proof point that small models can be useful. The "small specialized model" segment (Phi, Gemma, Llama 3.2-1B/3B, Qwen 0.5B/1.5B, Apple's on-device models) is now a real category, and Phi was the trigger.
Core Capabilities
Long Documents
Handles entire codebases, books, and multi-doc RAG.
Generative
Produces images, video, audio, or other media.
Context Window
128k tokens
≈ 98 pages
4k Chat 聊天
32k Long docs 长文档
128k This model 本模型
400k Multi-doc 多文档
1M Codebase 整个代码库
10M
Availability
API
Not available
Product / App
Not available
Open Source
Released
Enterprise
—
Pricing Model
Free / self-host
Open weights — pay only for compute.
Self-host Capability / Performance
Where this model sits relative to the middle 60% of models in the tree. All scores are 0–10 (higher is better).
Lower 20% Upper 80% This model
Context / memory
Context window size · log-scaled
6.0
9.0
6.0
Lower 20% 20th percentile — 20% of models score below this This model Where the current model lands Upper 80% 80th percentile — only 20% of models score above this
Percentile boundaries are computed across every model in the tree that reports the underlying benchmark for each capability.
What it feels like
- Phi-3-mini (3.8B) runs entirely on a phone yet matches GPT-3.5-class quality on common reasoning benchmarks
- First widely-deployed open SLM that proved high-quality synthetic + filtered data beats raw scale
- Outperforms models same-size and one-tier-up on language, reasoning, coding, and math benchmarks
- Available across Azure AI Studio, Hugging Face, and Ollama — meant for laptop / edge from day one
- Multilingual coverage is weaker than larger OSS models — primarily English-tuned
- Knowledge-heavy queries (history, trivia) clearly limited by parameter count
Best use cases
- On-device / phone / laptop deployments where no GPU is available
- Edge inference for IoT, mobile apps, offline assistants
- Reasoning-driven workflows where fine-tuned synthetic data beats general-knowledge breadth
- Distillation targets and benchmark baselines for SLM research
Tools to try
Not ideal for
- Broad-knowledge chat — limited factual recall by design
- Multilingual production beyond English
- Tasks where frontier-tier reasoning matters (use Phi-4 or larger models)
Model Evolution
phi is Microsoft's language model family.