Microsoft Phi-4 family
https://azure.microsoft.com/en-us/products/phi
Microsoft Research's Phi family of small, data-curation-focused open-weights models. Phi-4 (14B, Dec-2024) matches much larger models on STEM benchmarks. Phi-4 Multimodal (5.6B, Feb-2025) integrates speech + vision + text. MIT license. Substrate for on-device / edge agents and cost-sensitive OSS use.
Active · Verified 2026-10-31 · fresh as of 2026-10-31 · 41 cited facts, 1 with a third-party source · 2 recorded changes · Suggest a correction
At a glance
- Type
- Small / efficient open-weights model family (Phi-4 14B + Phi-4 multimodal + Phi-4 mini) vendor fix
- Tier
- T2
- Created
- 2023-06 (Phi-1); 2023-12 (Phi-2); 2024-04 (Phi-3); 2024-12 (Phi-4); 2025-02 (Phi-4 Multimodal) vendor fix
- Latest release
- Phi-4-reasoning-vision-15B (2026-03) third-party fix
- Pricing
- Free open-weights (MIT); Azure AI pay-per-token if hosted vendor fix
- Funding
- Microsoft (MSFT) public; ~$3T market cap; AI capex $80B+ 2025 vendor fix
- Freshness
- 2026-10-31 fix
- Layer
- large-models repo fix
- Licensing
- open-source vendor fix
- Maintainer health
- 0 issues in 90 days · 0 releases in 12 months · bus factor 2 — as of 2026-11-01 repo
When to use
Optimised for: on-device / edge inference; STEM reasoning at small footprint; data-curation-driven training
Anti-fit: not frontier-tier at general chat (smaller knowledge base); narrower world-knowledge than 70B+ models; multimodal less mature than Gemini 3
Pros & cons
Pros
MIT licensed (broadest permissions); 14B matches 70B+ on STEM; deployable on Copilot+ PCs / edge; Microsoft research pedigree.
Cons
Smaller world-knowledge than peers (data curation tradeoff); not at frontier-tier general capability; original Phi team leader (Bubeck) departed to OpenAI Oct-2024.
Claims & capabilities
Phi-4 14B matches Llama 3.3 70B on STEM benchmarks (MMLU 84.8%); Phi-4 Multimodal 5.6B integrates speech / vision / text; MIT license; data-curation-driven training methodology
Technical surface
- API surface
- HuggingFace Transformers; ONNX Runtime; Azure AI Foundry REST; native Windows ML
- Backend storage
- not applicable — substrate foundation model
- Deployment
- Both (self-hosted + managed cloud) — MIT open weights (Hugging Face, vLLM/Transformers); Microsoft Foundry catalog; on-device via ONNX Runtime
- Embedding model
- not applicable — not a memory product
- Multi-tenancy
- not applicable — substrate foundation model
- MCP
- not applicable — substrate foundation model
- A2A
- not applicable — substrate foundation model
- OpenTelemetry
- not applicable — substrate foundation model
Similar systems
Other foundation models (substrate reference) in the catalog, ranked by inbound references.
- Anthropic Claude (foundation models) T1
Claude family of frontier foundation models — Sonnet 4 (workhorse), Opus 4.5 (frontier capability), Haiku 4.5 (fast/cheap). Frontier model used as substrate by Claude Code, Anthropic Memory tool, Cline, Aider, Continue.dev, Cursor, Goose, and dozens of other catalog entries. Memory in Claude is parametric (in weights) — discrete episodic memory lives outside the model in the Claude Memory tool (see Platform-provider memory section).
- OpenAI GPT family (GPT-5 / GPT-4o / o3 / o4) T1
OpenAI's foundation model family — GPT-5 (2025-08 flagship), GPT-4o (multimodal workhorse), o3/o4 reasoning series. Substrate for ChatGPT, the OpenAI API, Microsoft Copilot, Azure OpenAI, and many downstream agents in this catalog (Devin, Codex CLI, Operator, Cursor mode). Parametric memory in weights; discrete user memory ships in ChatGPT Memory feature (separate row).
- Google Gemini 3 family T1
Google DeepMind's Gemini 3 frontier model family — Gemini 3 Pro (top tier), 3 Flash (fast), 3 Nano (on-device). Substrate for Google Workspace AI, Vertex AI Agent Platform, Gemini Code Assist, Project Mariner, and Project Astra. Multimodal-native; 1M-10M context window. Largest context window of any frontier model. Used by Gemini Enterprise Agent Platform Memory Bank (separate row).
- Mistral Large 2 / Mixtral family T1
French frontier-model lab (Paris). Family includes Mistral Large 2 (123B dense, Jul-2024), Mixtral 8x22B (open-weights MoE), Mistral Small 3 (24B, open-weights Jan-2025), Codestral (code specialist). EU-headquartered alternative to US/China model labs; GDPR-friendly data residency. Distinct from the Mistral Le Chat memory feature (already in catalog) — this row covers the model family itself as substrate.
- xAI Grok 4 T1
Elon Musk's xAI frontier model — Grok 4 launched 2025-07. Trained on Colossus (100k+ GPU supercluster in Memphis). Deeply integrated with X/Twitter (real-time tweet data access). Grok 4 Heavy variant for premium tier. No frontier-tier open-weights — Grok 1 (314B MoE) is the only weight release. The x.ai site now brands itself SpaceXSI (page title, checked 2026-10-29).
- Cohere Command R+ / Command A T1
Toronto-based foundation-model lab (Aidan Gomez, ex-Google Brain 'Attention Is All You Need' co-author). Command family is enterprise / RAG-optimised: Command A (Mar-2025, 111B dense, 256k ctx), Command R+ (104B MoE), Command R7B (smallest tier). Announced merger with Aleph Alpha April-2026. Open weights via CC-BY-NC (non-commercial); Cohere-hosted for commercial.
Sources (40 cited facts, 1 third-party)
Each fact links to the page it came from; the badge says who published it (how to read sources).
- What it is Small / efficient open-weights model family (Phi-4 14B + Phi-4 multimodal + Phi-4 mini) vendor
- What it does & what's distinct Microsoft Research's Phi family of small, data-curation-focused open-weights models. Phi… vendor
- Claims / benefits Phi-4 14B matches Llama 3.3 70B on STEM benchmarks (MMLU 84.8%); Phi-4 Multimodal 5.6B i… vendor
- Modalities text, code (Phi-4); + image, speech (Phi-4 Multimodal) vendor
- Created 2023-06 (Phi-1); 2023-12 (Phi-2); 2024-04 (Phi-3); 2024-12 (Phi-4); 2025-02 (Phi-4 Multi… vendor
- Latest release Phi-4-reasoning-vision-15B (2026-03) third-party
- License MIT vendor
- GitHub github.com/microsoft/Phi-3CookBook ~3k+ stars + various HuggingFace repos vendor
- Mindshare Strong small-model OSS mindshare; default Windows Copilot+ on-device model; HuggingFace … vendor
- Funding Microsoft (MSFT) public; ~$3T market cap; AI capex $80B+ 2025 vendor
- Customers / scale Azure AI customers; on-device deployment (Windows 11 Copilot+ PCs); Azure AI Foundry cat… vendor
- Pricing Free open-weights (MIT); Azure AI pay-per-token if hosted vendor
- Compliance MIT license — deployer-controlled; Azure inherits Microsoft compliance posture (SOC 2 / … vendor
- Data handling Open-weights — deployer-controlled vendor
- Deployment Both (self-hosted + managed cloud) — MIT open weights (Hugging Face, vLLM/Transformers);… vendor
- HQ Redmond, US (Microsoft Research) vendor
- Founders / pedigree Microsoft Research (Phi team under Sébastien Bubeck — now ex-MSR, joined OpenAI Oct-2024… vendor
- Performance MMLU 84.8% (Phi-4 14B); GPQA 56% (Phi-4); HumanEval 82%; outperforms many 70B+ models on… vendor
- Code/weights release MIT — weights public on HuggingFace + Azure AI Foundry catalog vendor
- API surface HuggingFace Transformers; ONNX Runtime; Azure AI Foundry REST; native Windows ML vendor
- LLM lock open (MIT) vendor
- Runtimes any (HuggingFace Transformers, ONNX Runtime, llama.cpp, MLX, Ollama, Windows ML) vendor
- Validated verticals on-device / edge (Copilot+ PCs, mobile); cost-sensitive OSS; STEM-focused agents; Azure … vendor
- Time to running <5min via Ollama; <30min Azure AI Foundry; on-device via Windows ML on Copilot+ PCs vendor
- Anti-fit not frontier-tier at general chat (smaller knowledge base); narrower world-knowledge tha… vendor
- Optimised for on-device / edge inference; STEM reasoning at small footprint; data-curation-driven trai… vendor
- Adjacent infrastructure Azure AI Foundry; ONNX Runtime; Windows Copilot+ PCs; HuggingFace Transformers; Ollama vendor
- Pros MIT licensed (broadest permissions); 14B matches 70B+ on STEM; deployable on Copilot+ PC… vendor
- Cons Smaller world-knowledge than peers (data curation tradeoff); not at frontier-tier genera… vendor
- Project & sources azure.microsoft.com/en-us/blog/welcome-phi-4/ vendor
- cost-input-usd-per-mtok 0 vendor
- cost-output-usd-per-mtok 0 vendor
- cost-tier free vendor
- cost-pricing-model free-tier vendor
- cost-last-verified 2026-06-29 vendor
- capability-benchmark-sources livecodebench: 53.8% (Phi-4-reasoning, 2024-08 to 2025-01 window, 2025-04-30) https://ar… paper
- architecture-deployment-locus hybrid vendor
- licensing-bucket open-source vendor
- positioning-platform-vs-product platform vendor
- taxonomy-three-layer-rollup large-models repo
What changed
Changes to this record's lifecycle, licence, release, pricing and hosting facts, from the catalog's git history. Subscribe (Atom)
- 2026-10-29
- Homepage “https://azure.microsoft.com/en-us/blog/welcome-phi-4/” → “https://azure.microsoft.com/en-us/products/phi”
- 2026-10-26
- Latest release “Phi-4 Multimodal (2025-02); Phi-4 (2024-12)” → “Phi-4-reasoning-vision-15B (2026-03)” source
- Deployment “Self-hostable (MIT) + Azure AI Foundry hosted + on-device (Windows Copilot+ PCs, ONNX Runtime)” → “Both (self-hosted + managed cloud) — MIT open weights (Hugging Face, vLLM/Transformers); Microsoft Foundry catalog; on-device via ONNX Runtime” source
Related systems
References (1)
- Azure Machine Learning depends on at runtime — adjacent-infrastructure cell: Azure AI Foundry; ONNX Runtime; Windows Copilot+ PCs; HuggingFace Transformers; Ollama