微软（Microsoft）Microsoft　·　当前代current generation Phi-4-reasoning 系Phi-4-reasoning line（2026-08）

PhiPhi模型家族谱系卡model family lineage card

微软的小尺寸高质量路线：MIT、尺寸集中在 4B–15B、官方主打「合成数据教小模型」。当前代是 reasoning-vision 线（谱系上仍是 phi-4 后训练）。本站未收其量化对比——当前代 GGUF 生态太薄（见下）。Microsoft's small-but-high-quality line: MIT, sizes concentrated at 4B–15B, and the vendor's pitch is "teach small models with synthetic data". The current line is reasoning-vision (still a phi-4 post-train on the weight lineage). This site carries no quantisation comparison — the current generation's GGUF ecosystem is too thin (see below).

本站未收 Phi 的量化对比：当前代 Phi-4-reasoning-vision-15B 的 GGUF 生态实测仅约 2.9k 下载（2026-10-30 HF 复测），远低于本站 ~30 万门槛；老代 phi-4 / mini 生态虽大但不是当前代，本站不采老代（收录标准第 1 条）。No quantisation comparison on this site: the current Phi-4-reasoning-vision-15B GGUF ecosystem measured only ~2.9k downloads (HF search 2026-10-30), far below the ~300k threshold; the older phi-4 / mini ecosystems are larger but are not the current generation, and this site does not collect older generations (collection rule #1).

先说清楚：这个站的数据怎么来的First, where this site's data comes from

✅ 实测字段Measured fields

直接从 Hugging Face 公开 API 读取的客观数据：文件体积（字节）、上下文长度、架构、许可、下载量、更新时间。Objective data read straight from the Hugging Face public API: file sizes in bytes, context length, architecture, licence, downloads, last updated.

✅ 推导Derived

由实测字段按公开规则算出：每权重比特数（依 llama.cpp 命名标准）、档位归属、显存需求区间。推导规则在Computed from the measured fields by published rules: bits per weight (per the llama.cpp naming convention), tier assignment, VRAM requirement range. The rules are fully published in METHODOLOGY.md 完整公开，任何人可复算。 — anyone can recompute them.

❌ 我们不说的What we do not claim

哪个量化质量更高、哪个更快、哪个模型更强——这些需要实测，本站不跑 benchmark，所以不装作知道。凡我们给出的主观判断，一律显式标注「我们的口径」。Which quantisation is higher quality, which is faster, which model is stronger — these need measurement. This site does not run benchmarks, so it does not pretend to know. Every subjective judgement we do offer is explicitly marked as our own reading.

本站This site does 不下载、不托管任何模型权重文件not download or host any model weight files。所有数字都来自公开元数据，点击即可回到发布页核对。. Every figure comes from public metadata, and clicking it takes you back to the release page to verify. 采集时间Collected: 2026-10-30（数据快照，非实时） (a snapshot, not live)

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-20

出品属地Origin　国外出品foreign-made　出品方为微软（美国）。Made by Microsoft (US).
国内下载Direct access　国内可直连direct OK　ModelScope 有 microsoft 官方组织镜像（2026-10-20 实测 200）；Hugging Face 直连不可达。Official microsoft org mirror on ModelScope (measured 200); Hugging Face direct unreachable.　·　国内镜像入口domestic mirror
备案状态CAC filing　未见于备案名单not in filing list：公示名单未检索到微软相关条目（2026-10-20 检索 13 批附件）。No Microsoft entry found in the public registry (13 batch attachments searched on 2026-10-20).

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「国内可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise.

网信办《生成式人工智能服务已备案信息》滚动公告页（13 个批次附件 PDF，2026-10-20 全量下载检索）：CAC · filed generative-AI service information (rolling announcement page, 13 batch attachment PDFs, full download & search on 2026-10-20): https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm
ModelScope 官方组织镜像（2026-10-20 实测 200）：ModelScope official-org mirror (200 verified on 2026-10-20): https://www.modelscope.cn/models/microsoft/Phi-4-reasoning-vision-15B

代次Generations

Phi-4-reasoning-vision-15BPhi-4-reasoning-vision-15B2026-08

当前代：15B 视觉推理档（模型卡 base_model 指向 Phi-4-reasoning），MIT 许可；技术报告 arXiv:2511.19663。Current generation: the 15B vision-reasoning build (model card base_model points at Phi-4-reasoning), MIT licence; technical report arXiv:2511.19663.
官方页面 ↗official page ↗
Phi-4-reasoning / -plusPhi-4-reasoning / -plus2025-11

推理强化代：官方 README「finetuned from Phi-4 using supervised fine-tuning on a dataset of chain-of-thought traces and reinforcement learning」，上下文 32K。Reasoning-focused step: official README "finetuned from Phi-4 using supervised fine-tuning on a dataset of chain-of-thought traces and reinforcement learning", 32K context.
官方页面 ↗official page ↗
phi-4 / Phi-4-mini / Phi-4-multimodalphi-4 / Phi-4-mini / Phi-4-multimodal2024-12–2025-03

基座代：phi-4 为 14B 密集 / 16K 上下文（官方 README 口径），mini 3.8B 与 multimodal 同代放出；全部 MIT。Base-generation step: phi-4 is 14B dense / 16K context per the official README, with the 3.8B mini and multimodal builds released alongside; all MIT.
官方页面 ↗official page ↗

派生标记Derived from 按官方模型卡 base_model 声明per the official model-card base_model declaration

官方模型卡声明当前代 Phi-4-reasoning-vision-15B 的 base_model 是 Phi-4-reasoning，而 Phi-4-reasoning 又声明派生自 phi-4（SFT + RL 微调）——整条当前代线在权重谱系上是 phi-4 的后训练产物，不是从零预训练。The official model card declares the current-generation Phi-4-reasoning-vision-15B is based on Phi-4-reasoning, which in turn declares phi-4 as its base (SFT + RL fine-tuning) — the whole current line is a post-train lineage of phi-4, not trained from scratch.　基座base：microsoft/Phi-4-reasoning　出处 ↗source ↗

尺寸档Size tiers

端侧 mini 3.8BOn-device mini 3.8B
单卡 14B–15B（phi-4 密集 / reasoning-vision 15B）Single-card 14B–15B (phi-4 dense / reasoning-vision 15B)

许可Licences

mitMIT（phi-4 与 reasoning 系模型卡均 license: mit，许可原文在仓内 LICENSE）MIT (phi-4 and the reasoning-line model cards all say license: mit; licence text in-repo) 许可原文 ↗licence text ↗

官方能力声明（带出处）Official capability claims (with sources)

官方 README（phi-4）：「14B parameters, dense decoder-only Transformer」「Context length: 16K tokens」；训练数据以合成数据集为主（官方明说 synthetic datasets 路线）。Official README (phi-4): "14B parameters, dense decoder-only Transformer"; "Context length: 16K tokens"; training leans on synthetic datasets (stated by the vendor).出处 ↗source ↗
官方 README（Phi-4-reasoning）：「state-of-the-art open-weight reasoning model finetuned from Phi-4 using supervised fine-tuning … and reinforcement learning」「Context length: 32k tokens」。Official README (Phi-4-reasoning): "state-of-the-art open-weight reasoning model finetuned from Phi-4 using supervised fine-tuning … and reinforcement learning"; "Context length: 32k tokens".出处 ↗source ↗

以上均为厂商官方页面原文或官方自评基准的转述；涉及横评结论的，本站不背书——本站不跑 benchmark。The above are quotations of official pages or transcriptions of vendor-run benchmarks; this site does not endorse cross-model conclusions — it runs no benchmarks.

出处Sources

phi-4 模型页与 README（2026-10-20 直抓 200）：phi-4 model page & README (direct fetch 200 on 2026-10-20): https://huggingface.co/microsoft/phi-4
Phi-4-reasoning 模型页与 README（2026-10-20 直抓 200，含 base_model 声明）：Phi-4-reasoning model page & README (direct fetch 200 on 2026-10-20, includes base_model declaration): https://huggingface.co/microsoft/Phi-4-reasoning
Phi-4-reasoning-vision-15B 模型页与 README（2026-10-20 直抓 200）：Phi-4-reasoning-vision-15B model page & README (direct fetch 200 on 2026-10-20): https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B

核验快照：2026-10-30 · 本站不实测、不排名，未知就写未知。Verification snapshot: 2026-10-30 · no benchmarks, no rankings — unknown stays unknown.