下载或取得权重
Phi-4
open_weights
获取入口：Phi-4 model card ↗选择要核对的信息，直接进入对应章节；各项事实均保留版本、条件和来源。
7/14 项有公开值 · 当前版本 Phi-4 · 数据截止 2026-10-20
页面只展示已审核 Claim；待审核数据不会被当作公开事实。只展示已审核事实；未公开、冲突和过期状态不会被补猜。
只判断选型资料是否足够，不评价模型能力高低。
所有必填事实均通过状态、来源、证据与时效校验。最早核验时间：2026-10-20。
release_datelifecycle_statuscapability_tagsdeployment_optionscontext_window_tokensweights_availabilitymodel_licensetotal_parametersPhi-4 是 Microsoft AI 的 Phi-4 模型，年鉴当前记录 1 个版本，最新公开记录为 Phi-4。
权重为开放权重，代码为未披露，训练数据为partially_disclosed，部署方式为可自行部署；这些维度分别判断，不能合并成单一“开源/闭源”结论。
当前公开快照的上下文窗口为 16.4K tokens，架构为 未公开；这里只使用通过审核且来源允许公开采集的 Claim。
7 个维度暂无可公开投影的审核记录
补充来源或线索会进入投稿队列，经编辑核验后才会成为公开事实。当前来源记录 1 个。
技术事实、业务价值、企业治理和证据质量分区呈现
厂商、家族、API ID、日期与生命周期
Dense/MoE、参数、层数、专家与 Tokenizer
预训练、SFT、偏好优化与安全对齐
教师/学生、基础模型、轨迹和限制
权重、代码、数据、许可证、部署与 API
输入输出、缓存、Batch 与部署成本边界
数据、权限、审计、合规、安全与人工兜底
8 条已审核事实 · 全部来自公开来源闸门通过的记录
尚无已审核公开记录的维度：02 能力与任务、07 部署与接入、09 性能与评测、11 业务场景、12 落地与提效。查看资料缺口 ↑
关系只来自年鉴数据库中已登记的归属字段。
当前最新版本的 approved Claim 投影 · 4 条历史事实正在复核
下载、API、客户端和更新内容只展示已登记的来源事实；未登记不作推断。
open_weights
获取入口：Phi-4 model card ↗self_hostable、transformers
接入说明：Phi-4 model card ↗保留发布日期精度、生命周期节点和原始来源，未公开日期不会被推断。
模型版本 · active
Microsoft published Phi-4 as a 14B MIT-licensed open-weight model.
来源：Phi-4 model card ↗显示最新核验的 8 条 · 共 8 条已审核事实
Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
1 个版本 · 每项可回到来源与技术解释
高置信度，已绑定来源与证据。
条件：Set on 2026-10-20 from the model card: intended use lists reasoning and logic, the training section emphasises math, coding and common-sense reasoning data, and the repository is tagged math and code.
来源：Phi-4 model card ↗Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
高置信度，已绑定来源与证据。
条件：Re-verified on 2026-10-20 against the current model card; Context length: 16K tokens (no extended-context configuration is documented).
来源：Phi-4 model card ↗Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
引用文本与接口返回都只包含已审核公开事实
Microsoft AI Phi-4｜最新版本 Phi-4｜已审核事实 8 条｜公开来源 1 个｜数据截止 2026-10-20｜AI 年鉴｜https://ainianjian.com/models/model.microsoft.phi4
公开接口返回该模型的版本、已审核事实 8 条及其来源信息，可直接用于脚本或对比。
https://api.ainianjian.com/api/models/model.microsoft.phi4 ↗curl -s 'https://api.ainianjian.com/api/models/model.microsoft.phi4'
技术规格使用当前版本已审核且来源允许公开采集的事实；模型级开放信息与跨版本地区记录分别标注范围。请按每项适用范围核对，未登记字段显示「未公开」，不使用估算值填充。
Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
高置信度，已绑定来源与证据。
条件：Re-verified on 2026-10-20; the card documents local inference through transformers and the repository is tagged text-generation-inference and endpoints_compatible. No managed first-party API price is recorded here.
来源：Phi-4 model card ↗Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
高置信度，已绑定来源与证据。
条件：Set on 2026-10-20 because the microsoft/phi-4 repository remains public, not gated and not disabled; Microsoft does not publish a formal lifecycle table for this release, so this reflects current availability rather than a vendor-stated status.
来源：Phi-4 model card ↗Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
高置信度，已绑定来源与证据。
条件：Re-verified on 2026-10-20; the model card Model Summary table lists License: MIT and the front matter declares license: mit.
来源：Phi-4 model card ↗Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
高置信度，已绑定来源与证据。
条件：Set on 2026-10-20 from the model card Model Summary: Release date: December 12, 2024.
来源：Phi-4 model card ↗Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
高置信度，已绑定来源与证据。
条件：Re-verified on 2026-10-20; Architecture: 14B parameters, dense decoder-only Transformer model.
来源：Phi-4 model card ↗Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.
高置信度，已绑定来源与证据。
条件：Re-verified on 2026-10-20; the official microsoft/phi-4 Hugging Face repository is public, not gated and exposes safetensors weights.
来源：Phi-4 model card ↗Model Summary: phi-4 is a 14B parameter dense decoder-only Transformer; Context length: 16K tokens; GPUs: 1920 H100-80G; Training data: 9.8T tokens; Release date: December 12, 2024; License: MIT; Status: static model trained on an offline dataset with cutoff dates of June 2024. Intended use highlights memory/compute constrained environments, latency bound scenarios, and reasoning and logic. Front matter: license: mit, tags include math, code, chat, conversational; library_name: transformers. The Hugging Face model API reports the repository as public and not gated.