{"id":"microsoft-phi-4-family--azure-microsoft-com","name":"Microsoft Phi-4 family","url":"https://azure.microsoft.com/en-us/products/phi","page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/microsoft-phi-4-family--azure-microsoft-com","family":"models-ops","section":"Foundation models (substrate reference)","sections":["Foundation models (substrate reference)"],"tier":2,"last_verified_at":"2026-10-31","lifecycle":{"status":"active","severity":0,"note":null},"taxonomy":null,"cells":{"type":{"value":"Small / efficient open-weights model family (Phi-4 14B + Phi-4 multimodal + Phi-4 mini)","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"desc":{"value":"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.","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"claims":{"value":"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","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"modalities":{"value":"text, code (Phi-4); + image, speech (Phi-4 Multimodal)","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"created":{"value":"2023-06 (Phi-1); 2023-12 (Phi-2); 2024-04 (Phi-3); 2024-12 (Phi-4); 2025-02 (Phi-4 Multimodal)","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","last_verified_at":"2026-05-13","source_class":"vendor"},"latest-release":{"value":"Phi-4-reasoning-vision-15B (2026-03)","citation":"https://en.wikipedia.org/wiki/Phi_(language_model)","status":"real-data","tier":"T2","last_verified_at":"2026-10-26","source_class":"third-party"},"license":{"value":"MIT","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"gh":{"value":"github.com/microsoft/Phi-3CookBook ~3k+ stars + various HuggingFace repos","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","last_verified_at":"2026-05-13","source_class":"vendor"},"mindshare":{"value":"Strong small-model OSS mindshare; default Windows Copilot+ on-device model; HuggingFace top-10 small-model family by downloads","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","last_verified_at":"2026-05-13","source_class":"vendor"},"citations":{"value":"not applicable — not academic","citation":null,"status":"not-applicable","tier":"T3","last_verified_at":"2026-05-13"},"funding":{"value":"Microsoft (MSFT) public; ~$3T market cap; AI capex $80B+ 2025","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","last_verified_at":"2026-05-13","source_class":"vendor"},"customers":{"value":"Azure AI customers; on-device deployment (Windows 11 Copilot+ PCs); Azure AI Foundry catalog; HuggingFace OSS deployers","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"pricing":{"value":"Free open-weights (MIT); Azure AI pay-per-token if hosted","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"compliance":{"value":"MIT license — deployer-controlled; Azure inherits Microsoft compliance posture (SOC 2 / ISO / HIPAA / FedRAMP)","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"data-handling":{"value":"Open-weights — deployer-controlled","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"deployment":{"value":"Both (self-hosted + managed cloud) — MIT open weights (Hugging Face, vLLM/Transformers); Microsoft Foundry catalog; on-device via ONNX Runtime","citation":"https://labs.ai.azure.com/projects/phi-4-reasoning-vision-15b/","status":"real-data","tier":"T2","source_class":"vendor"},"hq":{"value":"Redmond, US (Microsoft Research)","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"founders":{"value":"Microsoft Research (Phi team under Sébastien Bubeck — now ex-MSR, joined OpenAI Oct-2024; Microsoft Research AI Frontiers org continues)","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"volume":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"perf":{"value":"MMLU 84.8% (Phi-4 14B); GPQA 56% (Phi-4); HumanEval 82%; outperforms many 70B+ models on STEM at 14B parameters","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"repro":{"value":"not applicable — not a research paper","citation":null,"status":"not-applicable","tier":"T3"},"code-release":{"value":"MIT — weights public on HuggingFace + Azure AI Foundry catalog","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"api-surface":{"value":"HuggingFace Transformers; ONNX Runtime; Azure AI Foundry REST; native Windows ML","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"latency":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"throughput":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"backend-storage":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"multi-tenancy":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"encryption":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"sso-rbac":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"embedding-model":{"value":"not applicable — not a memory product","citation":null,"status":"not-applicable","tier":"T3"},"consistency":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"versioning":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"tombstoning":{"value":"not applicable — not a memory product","citation":null,"status":"not-applicable","tier":"T3"},"schema-evolution":{"value":"not applicable — not a memory product","citation":null,"status":"not-applicable","tier":"T3"},"namespace":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"contradiction":{"value":"not applicable — not a memory product","citation":null,"status":"not-applicable","tier":"T3"},"forgetting":{"value":"not applicable — not a memory product","citation":null,"status":"not-applicable","tier":"T3"},"mcp-support":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"a2a-support":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"otel":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"webhooks":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"import-export":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"integration-count":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"orchestration":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"programmatic-control":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"vendor-benchmarks":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3","last_verified_at":"2026-05-13"},"pricing-specifics":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"agent-abstraction":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"memory-primitives":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"llm-lock":{"value":"open (MIT)","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"runtimes":{"value":"any (HuggingFace Transformers, ONNX Runtime, llama.cpp, MLX, Ollama, Windows ML)","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"session-handling":{"value":"not applicable — substrate foundation model","citation":null,"status":"not-applicable","tier":"T3"},"validated-verticals":{"value":"on-device / edge (Copilot+ PCs, mobile); cost-sensitive OSS; STEM-focused agents; Azure AI customers","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"time-to-running":{"value":"<5min via Ollama; <30min Azure AI Foundry; on-device via Windows ML on Copilot+ PCs","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"anti-fit":{"value":"not frontier-tier at general chat (smaller knowledge base); narrower world-knowledge than 70B+ models; multimodal less mature than Gemini 3","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"optimised-for":{"value":"on-device / edge inference; STEM reasoning at small footprint; data-curation-driven training","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"adjacent-infrastructure":{"value":"Azure AI Foundry; ONNX Runtime; Windows Copilot+ PCs; HuggingFace Transformers; Ollama","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"pros":{"value":"MIT licensed (broadest permissions); 14B matches 70B+ on STEM; deployable on Copilot+ PCs / edge; Microsoft research pedigree.","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"cons":{"value":"Smaller world-knowledge than peers (data curation tradeoff); not at frontier-tier general capability; original Phi team leader (Bubeck) departed to OpenAI Oct-2024.","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"links":{"value":"azure.microsoft.com/en-us/blog/welcome-phi-4/","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"obs-langsmith":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"obs-opentelemetry":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"obs-datadog":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"obs-helicone":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"obs-weave":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"obs-langfuse":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"obs-arize":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"obs-custom":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"cost-token-budget":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"cost-prompt-caching":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"cost-semantic-caching":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"cost-batching":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"cost-model-routing":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"cost-streaming-only":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"cost-observability-cost-attribution":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"eval-langsmith-evals":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"eval-braintrust":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"eval-weights-and-biases-agent":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"eval-helicone-evals":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"eval-custom-test-harness":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"eval-human-loop":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"eval-production-traffic-replay":{"value":"","citation":null,"status":"no-data","tier":"T3","last_verified_at":"2026-05-14"},"commit-trajectory":{"value":"not applicable — no GitHub repo","citation":null,"status":"not-applicable","tier":"T3","last_verified_at":"2026-05-14"},"citation-trajectory":{"value":"not applicable — non-paper row","citation":null,"status":"not-applicable","tier":"T3","last_verified_at":"2026-05-14"},"download-trajectory":{"value":"not applicable — not published as a package","citation":null,"status":"not-applicable","tier":"T3","last_verified_at":"2026-05-14"},"cost-input-usd-per-mtok":{"value":"0","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"cost-output-usd-per-mtok":{"value":"0","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"cost-tier":{"value":"free","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"cost-pricing-model":{"value":"free-tier","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"cost-last-verified":{"value":"2026-06-29","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"capability-composite-score":{"value":"65","citation":null,"status":"estimate","tier":"T3"},"capability-band":{"value":"competent","citation":null,"status":"estimate","tier":"T3"},"capability-benchmark-sources":{"value":"livecodebench: 53.8% (Phi-4-reasoning, 2024-08 to 2025-01 window, 2025-04-30) https://arxiv.org/abs/2504.21318 · gpqa: 69.3% (Phi-4-reasoning-plus, GPQA Diamond, 2025-04-30) https://arxiv.org/abs/2504.21318 · aime: 81.3% (Phi-4-reasoning-plus, AIME 2024, 2025-04-30) https://arxiv.org/abs/2504.21318","citation":"https://arxiv.org/abs/2504.21318","status":"real-data","tier":"T2","source_class":"paper"},"capability-last-verified":{"value":"2026-06-29","citation":null,"status":"estimate","tier":"T3"},"use-case-tags":{"value":"scoped-agentic, memory-augmented-chat, offline-capable","citation":null,"status":"estimate","tier":"T3"},"use-case-anti-tags":{"value":"not applicable — no anti-tags apply","citation":null,"status":"not-applicable","tier":"T3"},"adoption-accelerator":{"value":"","citation":null,"status":"no-data","tier":"T3"},"architecture-deployment-locus":{"value":"hybrid","citation":"https://azure.microsoft.com/en-us/products/phi","status":"real-data","tier":"T2","source_class":"vendor"},"licensing-bucket":{"value":"open-source","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","status":"real-data","tier":"T2","source_class":"vendor"},"positioning-platform-vs-product":{"value":"platform","citation":"https://azure.microsoft.com/en-us/products/phi","status":"real-data","tier":"T2","source_class":"vendor"},"taxonomy-three-layer-rollup":{"value":"large-models","citation":"https://github.com/MrPeppersDev/agent-infrastructure-landscape/blob/main/docs/schema/section-to-three-layer-rollup.json","status":"real-data","tier":"T1","source_class":"repo"},"developer-activity-opendigger":{"value":"","citation":null,"status":"no-data","tier":"T3"},"freshness-date":{"value":"2026-10-31","citation":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/microsoft-phi-4-family--azure-microsoft-com","status":"real-data","tier":"T2","source_class":"catalog"},"compliance-certifications":{"value":"","citation":null,"status":"no-data","tier":"T3"},"sdk-languages":{"value":"","citation":null,"status":"no-data","tier":"T3"},"protocol-support":{"value":"","citation":null,"status":"no-data","tier":"T3"},"maturity":{"value":"","citation":null,"status":"no-data","tier":"T3"}},"corroboration":{"cited_cells":40,"third_party_cells":1},"edges":{"outgoing":[{"source":"microsoft-phi-4-family--azure-microsoft-com","target":"azure-machine-learning--azure-microsoft-com","type":"runtime-dependency","evidence":"adjacent-infrastructure cell: Azure AI Foundry; ONNX Runtime; Windows Copilot+ PCs; HuggingFace Transformers; Ollama","citation":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/"}],"incoming":[]},"history":[{"date":"2026-10-29","commit":"a5ff184341","pr":305,"kind":"update","changes":[{"field":"url","label":"Homepage","from":"https://azure.microsoft.com/en-us/blog/welcome-phi-4/","to":"https://azure.microsoft.com/en-us/products/phi"}]},{"date":"2026-10-26","commit":"80cd27834a","pr":258,"kind":"update","changes":[{"field":"latest-release","label":"Latest release","from":"Phi-4 Multimodal (2025-02); Phi-4 (2024-12)","to":"Phi-4-reasoning-vision-15B (2026-03)","source":"https://en.wikipedia.org/wiki/Phi_(language_model)"},{"field":"deployment","label":"Deployment","from":"Self-hostable (MIT) + Azure AI Foundry hosted + on-device (Windows Copilot+ PCs, ONNX Runtime)","to":"Both (self-hosted + managed cloud) — MIT open weights (Hugging Face, vLLM/Transformers); Microsoft Foundry catalog; on-device via ONNX Runtime","source":"https://labs.ai.azure.com/projects/phi-4-reasoning-vision-15b/"}]}],"license":"CC-BY-4.0","source":"https://github.com/MrPeppersDev/agent-infrastructure-landscape/blob/main/data/landscape.json"}