Microsoft (Phi) models

Microsoft's Phi family — small models tuned for reasoning. The honest, $0-markup cost to run each — free on your machine, a rented GPU, or your own API key.

Phi-4 — 14B, Microsoft · $0.21/1M API
Phi-3.5-mini Instruct — 3.8B, Microsoft · $0.13/1M API
Phi 4 mini instruct — 3.8B, microsoft · $0.13/1M API
Phi tiny MoE instruct — 3.8B, microsoft · $0.13/1M API
Phi 3 mini 4k instruct — 3.8B, microsoft · $0.13/1M API
phi 2 — 2.8B, microsoft · $0.12/1M API
Phi 3 vision 128k instruct — 4.1B, microsoft · $0.13/1M API
Phi 3 mini 128k instruct — 3.8B, microsoft · $0.13/1M API
Phi mini MoE instruct — 7.6B, microsoft · $0.16/1M API
Phi 3.5 MoE instruct — 41.9B, microsoft · $0.44/1M API
Phi 4 mini reasoning — 3.8B, microsoft · $0.13/1M API
phi 1 5 — 1.4B, microsoft · $0.11/1M API

Other publishers

Qwen (Alibaba) · Meta Llama · DeepSeek · Google (Gemma) · Mistral AI · NVIDIA · OpenAI (gpt-oss) · IBM Granite · Z.ai (GLM) · Moonshot AI (Kimi) · MiniMax · Cohere · Ai2 (OLMo) · Nous Research · Liquid AI · TII (Falcon) · EleutherAI · InternLM · Xiaomi (MiMo)

All models → · Compare →