跳转至主内容 了解 Azure Microsoft 即零号客户 查看所有产品 (200+) Microsoft Foundry Azure Copilot GitHub Copilot Azure Kubernetes 服务 (AKS) Azure Cosmos DB Azure Database for PostgreSQL Azure Arc Microsoft Fabric Azure 上的 Linux 虚拟机 Foundry 模型 Foundry 智能体服务 Foundry IQ Foundry Tools Foundry Control Plane Foundry Control Plane 中的可观测性 Foundry 模型中的 Azure OpenAI Foundry Tools 中的 Azure 语音 Azure 机器学习 查看所有数据库 Azure Cosmos DB Azure DocumentDB Azure SQL Azure Database for PostgreSQL Azure 托管的 Redis Microsoft Fabric Azure Databricks Azure 上的 Linux 虚拟机 Azure 上的 Windows Server Azure Functions Azure 虚拟机规模集 Azure API 管理 Azure 容器应用 Azure Kubernetes 服务 (AKS) Azure Kubernetes 舰队管理器 Azure 容器注册表 Azure Red Hat OpenShift Azure 容器实例 Azure 容器存储 Azure Arc Azure Local Microsoft Defender for Cloud Azure Monitor Microsoft Sentinel Azure Migrate 查看所有解决方案 (40+) 适用于中小型企业的云解决方案 云迁移和现代化中心 AI 数据分析 Azure 数据库 AI 应用和智能体 Microsoft 市场 Microsoft Sovereign Cloud AI 应用和智能体 Azure 的负责任 AI AI 基础结构 AI 数据分析 机器学习运营 (MLOps) Azure 上的低代码应用程序开发 集成服务 无服务器计算 DevOps 迁移和现代化中心 .NET 应用迁移 Azure 上的数据库 Linux on Azure Azure 上的 Oracle Microsoft 云上的 SAP 自适应云 高性能计算 (HPC) 基础结构即服务 (IaaS) 复原能力 Azure Essentials 适用于 Azure 的 Frontier Accelerate Azure 上的 FinOps Microsoft 市场 Azure 定价概述 创建 Azure 帐户 免费 Azure 服务 灵活的购买选项 定价计算器 Azure 上的 FinOps 最大限度地提高 AI 的 ROI Azure 节省计划 Azure 预留 Azure 混合权益 虚拟机 Azure SQL Microsoft Foundry Microsoft Fabric Azure Kubernetes 服务 (AKS) Microsoft Defender for Cloud 查看更多 软件开发公司 Microsoft 市场 寻找合作伙伴 面向 Azure 合作伙伴的资源 开始使用 Azure 客户案例 分析师报告、白皮书和电子书 视频 详细了解云计算 文件 探索 Azure 门户 开发人员资源 快速启动模板 面向初创企业的资源 开发者社区 学生 面向合作伙伴的 Azure 部落格 活动与网络研讨会 了解 支援 连络销售人员 开始使用 Azure 登录

Azure OpenAI 服务定价

Azure OpenAI 服务定价概述

Azure OpenAI Service delivers enterprise-ready generative AI featuring powerful models from OpenAI, enabling organizations to innovate with text, audio, and vision capabilities. Beyond the cutting-edge models, companies choose Azure OpenAI Service for built-in data privacy, regional/area/global flexibility, and seamless integration into the Azure ecosystem including Fabric, Cosmos DB and Azure AI Search. Companies of all sizes can confidently scale AI solutions to enhance customer experience, automate workflows, and unlock creative potential, driving measurable impact and competitive differentiation.

To help customers in the journey, we offer pricing and cost management solutions to meet your needs. including:
  • Standard (On-Demand): Pay-as-you-go for input and output tokens.
  • Provisioned (PTUs): Allocate throughput with predictable costs, with monthly and annual reservations available to reduce overall spend.
  • Batch API: Language models are also now available in the Batch API for global deployments and three regions, that returns completions within 24 hours for a 50% discount on Global Standard Pricing.
You can choose from the following deployment types for Standard and Provisioned, which enable greater flexibility and control of pricing and performance. This flexibility helps when there is increasingly more restrictive data processing boundaries and need for increased throughput and lower price.
  • Global Deployment – Global SKU
  • Data Zone Deployment – Geographic based (EU or US)
  • Regional Deployment – Local Region (up to 27 regions)

浏览定价选项

应用筛选器来根据你的需求自定义定价选项。

价格仅是估算值,不应用作实际报价单。实际定价可能因与 Microsoft 签订的协议类型、购买日期和货币汇率而异。价格基于美元计算,并使用在上一个月末最后一个工作日之前的两个工作日内捕获的伦敦收盘即期汇率进行转换。如果本月底之前的两个工作日在主要市场中属于银行假日,则汇率确定日通常是紧接在此两个工作日之后的一天。此汇率适用于下个月的所有交易。登录到 Azure 定价计算器查看基于你与 Microsoft 的当前计划/产品/服务的定价。有关定价或请求报价的详细信息,请联系 Azure 销售专家。请参阅经常询问有关 Azure 定价的问题。

GPT-Chat Latest

模型 定价(每 100 万个令牌)
GPT-Chat Latest 10142026 Global 输入: $-
缓存输入: $-
输出: $-
GPT-Chat Latest 10142026 输入: $-
缓存输入: $-
输出: $-

GPT-6.1 Series

模型 定价(每 100 万个令牌) Priority Processing (1M Tokens)
GPT-6.1 Sol (short context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A
GPT-6.1 Sol (short context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A
GPT-6.1 Sol (long context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A
GPT-6.1 Sol (long context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A
GPT-6.1 Terra Short Context Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A
GPT-6.1 Terra Short Context Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A
GPT-6.1 Terra Long Context Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A
GPT-6.1 Terra Long Context Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A

GPT-6 Series

GPT-6 Astra, OpenAI's newest frontier model, is built to decide for organization, it reasons before it acts, commits to a course of action, and executes across the applications customers already run. Computer use is the marquee capability of this release.

模型 定价(每 100 万个令牌) Flex Pricing (1M Tokens) Priority Processing (1M Tokens)
GPT-6 Astra (short context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A N/A
GPT-6 Astra (short context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A N/A
GPT-6 Astra (long context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A
GPT-6 Astra (long context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A N/A
GPT-6 Sol (short context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-6 Sol (short context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-6 Sol (long context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-6 Sol (long context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-6 Luna (short context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A N/A
GPT-6 Luna (short context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A N/A
GPT-6 Luna (long context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A N/A
GPT-6 Luna (long context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A N/A

GPT-5.6 Series

模型 定价(每 100 万个令牌) Flex Pricing (1M Tokens) Priority Processing (1M Tokens)
GPT-5.6-sol (short context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-5.6-sol (short context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-5.6-sol (long context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-5.6-sol (long context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-5.6-terra (short context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-5.6-terra (short context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-5.6-terra (long context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-5.6-terra (long context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-5.6-luna (short context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A
GPT-5.6-luna (short context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A N/A
GPT-5.6-luna (long context) Global 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
GPT-5.6-luna (long context) Data Zone 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-
N/A 输入: $-
缓存输入: $-
Cache writes: $-
输出: $-

GPT 5.5 Series

GPT-5.5 delivers advanced reasoning, instruction following, and agentic capabilities for production AI workloads.

模型 定价(每 100 万个令牌) Priority Processing (1M Tokens)
GPT-5.5 Global 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
GPT-5.5 Data Zone 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
GPT-5.5 Long Context Global 输入: $-
缓存输入: $-
输出: $-
N/A
GPT-5.5 Long Context Data Zone 输入: $-
缓存输入: $-
输出: $-
N/A

GPT 5.4 Series

Built for Reliable AI Production: Stronger reasoning, dependable execution, and agentic workflows at scale.

模型 定价(每 100 万个令牌) Priority Processing (1M Tokens) 批处理 API (100 万个令牌)的定价
GPT-5.4 (<272k context length) Global 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
GPT-5.4 (<272k context length) Data Zone 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
N/A
GPT-5.4 (>272k context length) Global 输入: $-
缓存输入: $-
输出: $-
N/A N/A
GPT-5.4 Pro (<272k context length) Global 输入: $-
输出: $-
N/A N/A
GPT-5.4 Pro (>272k context length) Global 输入: $-
输出: $-
N/A N/A
GPT-5.4 mini Global 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
GPT-5.4 mini Data Zone 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
N/A
GPT-5.4 nano Global 输入: $-
缓存输入: $-
输出: $-
N/A N/A

GPT-5.3 Series

Unifies the frontier coding performance of GPT-5.2-Codex with the reasoning and professional knowledge capabilities of GPT-5.2.

模型 定价(每 100 万个令牌) Priority Processing (1M Tokens)
GPT-5.3 Codex Global 输入: $-
缓存输入: $-
输出: $-
N/A
GPT-5.3 Chat Global 输入: $-
缓存输入: $-
输出: $-
N/A

GPT-5.2

GPT-5.2 delivers the deep reasoning and expanded context handling necessary for building sophisticated AI agents capable of automating complex, long-running tasks across all business functions.

模型 定价(每 100 万个令牌) Priority Processing (1M Tokens)
GPT-5.2 Codex Global 输入: $-
缓存输入: $-
输出: $-
N/A
GPT-5.2 Global 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
GPT-5.2 Data Zone 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
GPT-5.2-chat latest Global 输入: $-
缓存输入: $-
输出: $-
N/A
GPT-5.2-chat Data Zone 输入: $-
缓存输入: $-
输出: $-
N/A

GPT-5.1

The GPT-5.1 series is built to respond faster to users in a variety of situations with adaptive reasoning, improving latency and cost efficiency across the series by varying thinking time more significantly. This, combined with other tooling improvements, enhanced stepwise reasoning visibility, multimodal intelligence, and enterprise-grade compliance.

模型 定价(每 100 万个令牌) Priority Processing (1M Tokens)
GPT-5.1 Global 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
GPT-5.1 Data Zone 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
GPT-5.1-chat Global 输入: $-
缓存输入: $-
输出: $-
N/A
GPT-5.1-codex Global 输入: $-
缓存输入: $-
输出: $-
N/A
GPT-5.1-codex-max Global 输入: $-
缓存输入: $-
输出: $-
N/A
GPT-5.1-codex-mini Global 输入: $-
缓存输入: $-
输出: $-
N/A

GPT-5 系列

模型 定价(每 100 万个令牌) Priority Processing (1M Tokens) 批处理 API (100 万个令牌)的定价
GPT-5 2025-08-07 全球版 输入: $-
缓存输入: $-
输出: $-
N/A 输入: $-
缓存输入: $-
输出: $-
GPT-5 数据区域 输入: $-
缓存输入: $-
输出: $-
N/A 输入: $-
缓存输入: $-
输出: $-
GPT-5 Pro Global 输入: $-
输出: $-
N/A N/A
GPT-5 Codex Global 输入: $-
缓存输入: $-
输出: $-
N/A N/A
GPT-5-mini 全球版 输入: $-
缓存输入: $-
输出: $-
N/A N/A
GPT-5-mini 数据区域 输入: $-
缓存输入: $-
输出: $-
N/A N/A
GPT-5-nano 全球版 输入: $-
缓存输入: $-
输出: $-
N/A N/A
GPT-5-nano 数据区域 输入: $-
缓存输入: $-
输出: $-
N/A N/A
GPT-5 聊天全球版 输入: $-
缓存输入: $-
输出: $-
N/A N/A

深度研究

通过深度研究,开发人员和企业能够使用结构化且引文丰富的答案来自动执行复杂的研究任务。它适用于构建客户支持机器人、内部知识助手或市场分析工具。深度研究提供基于实时 Web 数据的透明可审核见解。搜索上下文令牌按所使用模型的输入令牌价格收费。对于使用必应搜索的基础设置以及用于阐明问题的基本 GPT 模型,将会向你单独收费。

模型 定价
o3 深度研究全球 输入: $-
缓存输入: $-
输出: $-

o3

o3 是 o 系列推理模型中具有强大功能的推理模型,拓展了编码、数学、科学和视觉感知等领域的发展前沿。它在需要进行多方面分析的复杂查询方面表现出色,并在分析图像、图表和图形等视觉任务中具有强劲的表现。该模型具有 20 万个令牌上下文窗口,其知识截止日期为 2024 年 6 月。

模型 定价(每 100 万个令牌) 批处理 API (100 万个令牌)的定价
o3 2025-04-16 全球 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
o3 2025-04-16 数据区域 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
o3 2025-04-16 区域 输入: $-
缓存输入: $-
输出: $-
N/A

o4-mini

o4-mini 是 OpenAI o 系列中的一款紧凑且经济高效的推理模型。它擅长数学、编码和可视化任务。该模型具有 20 万个令牌上下文窗口,其知识截止日期为 2024 年 6 月。

模型 定价(每 100 万个令牌) 批处理 API (100 万个令牌)的定价
o4-mini 2025-04-16 全球 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
o4-mini 2025-04-16 数据区域 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
o4-mini 2025-04-16 区域 输入: $-
缓存输入: $-
输出: $-
N/A

GPT-4.1 系列

GPT-4.1 系列是一种高度先进的通用模型,拥有广泛的世界知识和增强的理解用户意图的能力,这使其尤其擅长执行创造性任务和智能体式规划。该系列具有 100 万个令牌上下文窗口,其知识截止日期为 2024 年 6 月。

模型 Standard Pricing (1M Tokens) Priority Processing (1M Tokens) 批处理 API (100 万个令牌)的定价
GPT-4.1-2025-04-14 全球 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4.1-2025-04-14 数据区域 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4.1-2025-04-14 区域 输入: $-
缓存输入: $-
输出: $-
N/A N/A
GPT-4.1-mini-2025-04-14 全球 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4.1-mini-2025-04-14 数据区域 输入: $-
缓存输入: $-
输出: $-
输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4.1-mini-2025-04-14 区域 输入: $-
缓存输入: $-
输出: $-
N/A N/A
GPT-4.1-nano-2025-04-14 全球 输入: $-
缓存输入: $-
输出: $-
N/A 输入: $-
输出: $-
GPT-4.1-nano-2025-04-14 数据区域 输入: $-
缓存输入: $-
输出: $-
N/A 输入: $-
输出: $-
GPT-4.1-nano-2025-04-14 区域 输入: $-
缓存输入: $-
输出: $-
N/A N/A

Azure OpenAI 中的 Sora

Sora 是一种多模态的 AI 模型,现已在 Azure AI Foundry 中提供,旨在帮助创意团队通过无缝的 API 优先集成实现创意。它基于 Azure 的企业级基础结构,提供安全、可缩放的部署,用于将概念转换为高质量的可视化内容。

Sora 2

模型 Size: Output Resolution 每秒价格
Sora 2 Global Portrait: 720x1280
Landscape: 1280x720
$-

Sora

每秒价格 1-5s 6-10s 11-15s 16-20s
480 Square Global $- $- $- $-
480p Global $- $- $- $-
720 Square Global $- $- $- $-
720p Global $- $- $- $-
1080 Square Global $- $- $- $-
1080p Global $- $- $- $-

GPT-Image Series

GPT-Image models enhance DALL-E with better instruction following accurate text rendering, and support for image input and editing. The model is priced per token, with different pricing for text and image tokens.

模型 定价(每 100 万个令牌) 批处理 API (100 万个令牌)的定价
Image-2.5-flare Global 输入文本: $-
Cached Input Text: $-
输入图像: $-
Cached Input Image: $-
输出图像: $-
N/A
Image-2.5-sunburst Global 输入文本: $-
Cached Input Text: $-
输入图像: $-
Cached Input Image: $-
输出图像: $-
N/A
GPT-Image-2 Global 输入文本: $-
Cached Input Text: $-
输入图像: $-
Cached Input Image: $-
Output Text: $-
输出图像: $-
N/A
GPT-Image-1.5 Global 输入文本: $-
Cached Input Text: $-
输入图像: $-
Cached Input Image: $-
Output Text: $-
输出图像: $-
N/A
GPT-Image-1.5 Data Zone 输入文本: $-
Cached Input Text: $-
输入图像: $-
Cached Input Image: $-
Output Text: $-
输出图像: $-
N/A
GPT-Image-1-mini Global 输入文本: $-
Cached Input Text: $-
输入图像: $-
Cached Input Image: $-
输出图像: $-
N/A
GPT-Image-1 全球 输入文本: $-
Cached Input Text: $-
输入图像: $-
Cached Input Image: $-
输出图像: $-
N/A
GPT-Image-1 区域性 输入文本: $-
Cached Input Text: $-
输入图像: $-
Cached Input Image: $-
输出图像: $-
N/A
GPT-Image-1 数据可用区 输入文本: $-
输入图像: $-
输出图像: $-
N/A

GPT-4.5

GPT-4.5-preview 是最新的常规用途模型,具备深厚的世界知识和更好的用户意图理解能力,使其在创意任务和代理规划方面表现出色。该模型具有 128K 上下文,并且知识更新到 2023 年 10 月。

模型 定价(每 100 万个令牌) 批处理 API (100 万个令牌)的定价
GPT-4.5-Preview-2025-02-27 Global 输入: $-
缓存输入: $-
输出: $-
N/A

o1

o1 是面向复杂任务的新型推理模型系列。该模型具有 200K 上下文,并且知识更新到 2023 年 10 月。

模型 定价(每 100 万个令牌) 批处理 API (100 万个令牌)的定价
o1 2024-12-17 全球 输入: $-
缓存输入: $-
输出: $-
N/A
o1 2024-12-17 美国/欧盟 – 数据区域 输入: $-
缓存输入: $-
输出: $-
N/A
o1 2024-12-17 区域 输入: $-
缓存输入: $-
输出: $-
N/A
o1 预览版 2024-09-12 全球 输入: $-
缓存输入: $-
输出: $-
N/A
o1 预览版 2024-09-12 美国/欧洲 – 数据区域 输入: $-
缓存输入: $-
输出: $-
N/A
o1 预览版 2024-09-12 区域 输入: $-
缓存输入: $-
输出: $-
N/A

使用定价计算器制定计划

o3 Mini

o3 mini 是 o1 mini 模型的更新版本。o3-mini 是一个快速、经济高效的推理模型,专为编码、数学和科学应用场景而设计。

o3 mini 模型现在拥有一个可支持 20 万个词元的扩展上下文输入窗口,且最多可输出 10 万个词元,可通过充足的空间提供复杂且详细的响应。o1 mini 模型支持包含 12.8 万个词元的上下文输入。o3 和 o1 模型的知识截止时间都是 2023 年 10 月。

模型 定价(每 100 万个令牌) 批处理 API (100 万个令牌)的定价
o3 mini 2025-01-31 全球 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
o3 mini 2025-01-31-美国/欧洲 – 数据区域 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
o3 mini 2025-01-31 区域 输入: $-
缓存输入: $-
输出: $-
N/A
o1-mini 2024-09-12 全球 输入: $-
缓存输入: $-
输出: $-
N/A
o1-mini 2024-09-12 美国/欧洲 – 数据区域 输入: $-
缓存输入: $-
输出: $-
N/A
o1-mini 2024-09-12 区域 输入: $-
缓存输入: $-
输出: $-
N/A

使用定价计算器制定计划

Open Source Models

gpt-oss-120b offers a high-performing, open and controllable LLM—blending frontier reasoning skills with enterprise-grade flexibility and deployment autonomy.

模型 定价(每 100 万个令牌)
gpt-oss-120b 输入: $-
输出: $-

GPT-Live

模型 价格
GPT-Live Global $-/per 1 unit hour
GPT-Live Data Zone $-/per 1 unit hour

音频模型

现在,GPT-realtime 和 GPT-audio 模型已通过 Azure AI Foundry 和 Azure OpenAI 服务提供,支持生产级应用程序中的高保真、低延迟语音交互。其他音频模型包括 GPT-4o、transcribe mini 和 mini-tts,提供高级语音转文本和文本转语音功能,拥有情感丰富的声音、可定制的输出和卓越的准确性,非常适合实时客户呼叫中心、实时字幕和交互式语音智能体。这些模型利用预训练和蒸馏技术,以支持自然的轮流对话,并为多模态部署提供稳定 API。

模型 定价(每 100 万个令牌) 批处理 API (100 万个令牌)的定价
GPT-Realtime-2.1 Global 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

图像
输入: $-
缓存输入: $-
N/A
GPT-Realtime 2.1 Data Zone 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

图像
输入: $-
缓存输入: $-
N/A
GPT-Realtime 2.1 mini Global 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

图像
输入: $-
缓存输入: $-
N/A
GPT-Realtime 2.1 mini Data Zone 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

图像
输入: $-
缓存输入: $-
N/A
GPT-Realtime-2 Global 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

图像
输入: $-
缓存输入: $-
N/A
GPT-Realtime-Translate Global 输出: $-/小时 N/A
GPT-Realtime-Whisper Global 输出: $-/小时 N/A
GPT-realtime-1.5 Global 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

图像
输入: $-
缓存输入: $-
N/A
GPT-realtime-1.5 Data Zone 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

图像
输入: $-
缓存输入: $-
N/A
GPT-audio-1.5 Global 文本
输入: $-
输出: $-

音频
输入: $-
输出: $-

N/A
GPT-realtime 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

图像
输入: $-
缓存输入: $-
N/A
GPT-realtime-mini-2025-12-15 Global 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

图像
输入: $-
缓存输入: $-
N/A
GPT-audio 文本
输入: $-
输出: $-

音频
输入: $-
输出: $-
N/A
GPT-audio-mini-2025-12-15 Global 文本
输入: $-
输出: $-

音频
输入: $-
输出: $-
N/A
GPT-Transcribe Global $-/1 Unit Hour N/A
GPT-Live-Transcribe Global $-/1 Unit Hour N/A
GPT-4o-Transcribe 文本
输入: $-
输出: $-

音频
输入: $-
输出: N/A
N/A
GPT-4o-transcribe-diarize 文本
输入: $-
输出: $-

音频
输入: $-
输出: N/A
N/A
GPT-4o-mini-transcribe-2025-12-15 文本
输入: $-
输出: $-

音频
输入: $-
输出: N/A
N/A
GPT-4o-mini-TTS-2025-12-15 文本
输入: $-
输出: N/A

音频
输入: N/A
输出: $-
N/A

计算机使用代理(CUA)

计算机使用代理(CUA)是一种专用的 AI 模型,支持 AI 与图形用户界面(GUI)交互、导航应用程序,以及通过自然语言指令自动执行多步骤任务。CUA 模型可以作为响应 API 中的工具使用。

模型 定价
计算机使用预览版全局设置 输入: $- /100 万个令牌
输出: $- /100 万个令牌

内置工具

响应 API 和助手 API 支持与计算机使用、代码解释器、函数调用和文件搜索等工具无缝交互,使开发人员能够构建 AI 智能体。

工具 输入
计算机使用(仅限响应 API) 输入: $- /100 万个令牌
输出: $- /100 万个令牌
文件搜索工具调用(仅限响应 API) $-/1K 工具调用
文件搜索* $-/GB 矢量存储/天(1 GB 免费)
代码解释器** $-/会话

*GB 是指二进制吉字节,其中 1 gb 为 2^30 b。

**如果助手在两个不同的线程中同时调用代码解释器,这将创建两个代码解释器会话(2 * $-)。每个会话默认处于活动状态一小时,这意味着只有当用户在同一线程中持续向代码解释器提供说明达 1 小时时,你才需要支付此费用一次。

推理成本(输入和输出)因每个助手所用的 GPT 模型而异。如果助手在两个不同的线程中同时调用代码解释器,这将创建两个代码解释器会话(2 * $-)。每个会话默认处于活动状态一小时,这意味着价格为一小时内在同一线程中向代码解释器发出指令所产生的费用。

实时 API

作为实时 API 中的重要部分,GPT-4o-Realtime-Preview 支持多语言语音转语音功能。它针对实时、低延迟的对话进行了优化,可实现最低延迟的自然交互,非常适合聊天机器人和对话式 AI。GPT-4o 是专为复杂任务设计的全面且更强大的版本,而 GPT-4o Mini 则是更小、更实惠的选项,非常适合优先考虑成本效益和速度的更简单的应用程序。

模型 定价(每 100 万个令牌)
GPT-4o-Realtime-Preview-2024-12-17-Global 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-
GPT-4o-Realtime-Preview-2024-12-17-US/EU – 数据区域 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-
GPT-4o-Realtime-Preview-2024-12-17-Regional 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-
GPT-4o-Mini-Realtime-Preview-2024-12-17-Global 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-
GPT-4o-Mini-Realtime-Preview-2024-12-17-US/EU – 数据区域 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-
GPT-4o-Mini-Realtime-Preview-2024-12-17-Regional 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-
GPT-4o-Realtime-Preview-2024-10-01-Global 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-
GPT-4o-Realtime-Preview-2024-10-01-US/EU – 数据区域 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-
GPT-4o-Realtime-Preview-2024-10-01-Regional 文本
输入: $-
缓存输入: $-
输出: $-

音频
输入: $-
缓存输入: $-
输出: $-

聊天完成 API

作为聊天完成 API 的重要部分,GPT 4o-Audio-Preview 模型可处理和生成音频内容。它支持语音识别和音频合成等高级功能,非常适合异步语音交互和情绪分析。GPT-4o 是专为复杂任务设计的全面且更强大的版本,而 GPT-4o Mini 则是更小、更实惠的选项,非常适合优先考虑成本效益和速度的更简单的应用程序。

模型 定价(每 100 万个令牌)
GPT-4o-Audio-Preview-2024-12-17-Global 文本
输入: $-
输出: $-

音频
输入: $-
输出: $-
GPT-4o-Audio-Preview-2024-12-17-US/EU – 数据区域 文本
输入: $-
输出: $-

音频
输入: $-
输出: $-
GPT-4o-Audio-Preview-2024-12-17-Regional 文本
输入: $-
输出: $-

音频
输入: $-
输出: $-
GPT-4o-Mini-Audio-Preview-2024-12-17-Global 文本
输入: $-
输出: $-

音频
输入: $-
输出: $-
GPT-4o-Mini-Audio-Preview-2024-12-17-US/EU – 数据区域 文本
输入: $-
输出: $-

音频
输入: $-
输出: $-
GPT-4o-Mini-Audio-Preview-2024-12-17-Regional 文本
输入: $-
输出: $-

音频
输入: $-
输出: $-

GPT-4o

GPT-4o 是最先进的多模态模型,具有更强大的视觉能力,并且比 GPT-4 Turbo 更快、成本更低。该模型具有 128K 上下文,并且知识更新到 2023 年 10 月。

模型 定价(每 100 万个令牌) 批处理 API (100 万个令牌)的定价
GPT-4o-2024-1120 全球 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4o-2024-1120 美国/欧盟 – 数据区域 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4o-2024-1120 区域 输入: $-
缓存输入: $-
输出: $-
N/A
GPT-4o-2024-08-06 全球 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4o-2024-08-06 美国/欧盟 – 数据区域 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4o-2024-08-06 区域 输入: $-
缓存输入: $-
输出: $-
N/A
GPT-4o-2024-0513 全球 输入: $-
输出: $-
输入: $-
输出: $-
GPT-4o-2024-0513 美国/欧盟 – 数据区域 输入: $-
输出: $-
N/A
GPT-4o-2024-0513 区域 输入: $-
输出: $-
N/A

使用定价计算器制定计划

GPT-4o 微型

GPT-4o 微型是极为经济高效的小型模型,并且具备视觉功能。该模型具有 128K 上下文,并且知识更新到 2023 年 10 月。

模型 定价(每 100 万个令牌) 批处理 API (100 万个令牌)的定价
GPT-4o-mini-0718 全球 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4o-mini-0718 美国/欧洲 –数据区域 输入: $-
缓存输入: $-
输出: $-
输入: $-
输出: $-
GPT-4o-mini-0718 区域 输入: $-
缓存输入: $-
输出: $-
N/A

使用定价计算器制定计划

已预配

You can allocate and manage throughput for deployments, ensuring predictable performance and stable capacity. You are charged an hourly rate per model regardless of usage, but you can also secure additional savings through monthly and annual reservations. Discover how to transition your regional deployments and provisioned reservations to global and data zones on this Learn page. To understand if your desired model is available in your specific region for provisioned pricing please visit this Learn page or contact your local sales rep for more details.

模型 最小 PTU Per PTU Hourly pricing Per PTU Monthly Reservation Pricing Per PTU Yearly Reservation Pricing
GPT-6 Sol (short context) Global 15 $- $- $-
GPT-6 Sol (short context) Data Zone 15 $- $- $-
GPT-6 Astra (short context) Global 15 $- $- $-
GPT-6 Astra (short context) Data Zone 15 $- $- $-
GPT-5.6-sol (short context) Global 15 $- $- $-
GPT-5.6-sol (short context) Data Zone 15 $- $- $-
GPT-5.6 Sol (Short Context) Regional 50 $- $- $-
GPT-5.6-terra (short context) Global 15 $- $- $-
GPT-5.6-terra (short context) Data Zone 15 $- $- $-
GPT-5.6 Terra (Short Context) Regional 50 $- $- $-
GPT-5.5 Global 15 $- $- $-
GPT-5.5 Data Zone 15 $- $- $-
GPT-5.5 Regional 50 $- $- $-
GPT-5.4 (<272k context length) Global 15 $- $- $-
GPT-5.4 (<272k context length) Data Zone 15 $- $- $-
GPT-5.4 (<272k context length) Regional 50 $- $- $-
GPT-5.3 Codex Global 15 $- $- $-
GPT-5.3 Codex Data Zone 15 $- $- $-
GPT-5.3 Codex Regional 50 $- $- $-
GPT-5.2 Codex Global 15 $- $- $-
GPT-5.2 Codex Data Zone 15 $- $- $-
GPT-5.2 Codex Regional 50 $- $- $-
GPT-5.2 Global 15 $- $- $-
GPT-5.2 Data Zones 15 $- $- $-
GPT-5.2 Regional 50 $- $- $-
GPT-5.1 Codex Global 15 $- $- $-
GPT-5.1 Codex Data Zones 15 $- $- $-
GPT-5.1 Codex Regional 50 $- $- $-
GPT-5.1 Global 15 $- $- $-
GPT-5.1 Data Zones 15 $- $- $-
GPT-5.1 Regional 50 $- $- $-
GPT-5-mini 全球版 15 $- $- $-
GPT-5-mini Data Zones 15 $- $- $-
GPT-5-mini Regional 50 $- $- $-
GPT-5 全局 15 $- $- $-
GPT-5 数据区域 15 $- $- $-
GPT-5 区域 50 $- $- $-
GPT-4.1 全局 15 $- $- $-
GPT-4.1 数据区域 15 $- $- $-
GPT-4.1 区域 50 $- $- $-
GPT-4.1-mini 全局 15 $- $- $-
GPT-4.1-mini 美国/欧洲数据区域 15 $- $- $-
GPT-4.1-mini 区域 25 $- $- $-
GPT-4.1-nano 全局 15 $- $- $-
GPT-4.1-nano 美国/欧洲数据区域 15 $- $- $-
GPT-4.1-nano 区域 25 $- $- $-
o3-mini 全局 15 $- $- $-
o3-mini US/EU 数据区域 15 $- $- $-
o3-mini 区域 25 $- $- $-
o3 全局 15 $- $- $-
o3 美国/欧洲数据区域 15 $- $- $-
o3 区域 50 $- $- $-
o4-mini 全局 15 $- $- $-
o4-mini 美国/欧洲数据区域 15 $- $- $-
o4-mini 区域 25 $- $- $-
GPT-4o 全球 15 $- $- $-
GPT-4o US/EU 数据区域 15 $- $- $-
GPT-4o 区域性 50 $- $- $-
微调 GPT-4o-区域 50 $- $- $-
GPT-4o 微型全球 15 $- $- $-
GPT-4o 微型 US/EU 数据区域 15 $- $- $-
GPT-4o 微型区域性 25 $- $- $-
微调 GPT-4o-Mini 区域 25 $- $- $-

使用定价计算器制定计划

基本模型

模型 每 1,000 个令牌的使用情况
Babbage-002 $-
Davinci-002 $-

微调模型

模型 定价
o4-mini (强化微调) 区域性 输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $-/小时
托管: $-/小时

评分器输入:
GPT-4o: $- /100 万个令牌
o3-mini: $- /100 万个令牌

评分器缓存输入:
GPT-4o: $- /100 万个令牌
o3-mini: $- /100 万个令牌

评分器输出:
GPT-4o: $- /100 万个令牌
o3-mini: $- /100 万个令牌
全球 输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $-/小时
托管: $-/小时

评分器输入:
GPT-4o: $- /100 万个令牌
o3-mini: $- /100 万个令牌

评分器缓存输入:
GPT-4o: $- /100 万个令牌
o3-mini: $- /100 万个令牌

评分器输出:
GPT-4o: $- /100 万个令牌
o3-mini: $- /100 万个令牌
开发人员 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $-/小时
GPT-4.1 区域性 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
全球 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
开发人员 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
GPT-4.1-mini 区域性 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
全球 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
开发人员 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
GPT-4.1-nano 区域性 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
全球 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
开发人员 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
GPT-4o-2024-08-06 区域性 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
全球 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
开发人员 培训: $- /100 万个令牌
GPT-4o-mini 区域性 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
全球 输入: $- /100 万个令牌
缓存输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
开发人员 培训: $- /100 万个令牌
GPT-3.5-Turbo (16K) 区域性 输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
GPT OSS 20B 区域性 输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时
Data Zone 输入: $- /100 万个令牌
输出: $- /100 万个令牌
培训: $- /100 万个令牌
托管: $-/小时

图像模型

模型 质量 分辨率 价格(每 100 张图像)
Dall-E-3 标准 1024 * 1024 $-
标准 1024 * 1792,
1792 * 1024
$-
Dall-E-3 HD 1024 * 1024 $-
HD 1024 * 1792,
1792 * 1024
$-
Dall-E-2 标准 1024 * 1024 $-

嵌入模型

模型 每 1,000 个标记
Ada $-
Ada DataZone $-
文本-嵌入-3-大 $-
text-embedding-3-large DataZone $-
文本-嵌入-3-小 $-
text-embedding-3-small DataZone $-

语音模型

模型 价格
Whisper $-/小时
TTS(文本转语音) $-/1 百万个字符
TTS HD $-/1 百万个字符

旧版语言模型

模型 上下文 输入(每 100 万个令牌) 输出(每 100 万个令牌)
GPT-3.5-Turbo-0301 4K $- $-
GPT-3.5-Turbo-0613 4K $- $-
GPT-3.5-Turbo-0613 16K $- $-
GPT-3.5-Turbo-1106 16K $- $-
GPT-3.5-Turbo-0125 16K $- $-
GPT-3.5-Turbo-Instruct 4K $- $-
GPT-4-Turbo 128K $- $-
GPT-4-Turbo-Vision 128K $- $-
GPT-4 8K $- $-
GPT-4 32K $- $-

Azure 定价和购买选项

直接与我们联系

获取 Azure 定价演练。了解云解决方案的定价、学习成本优化和请求自定义建议。

与销售专家交谈

查看购买方式

通过 Azure 网站、Microsoft 代表或 Azure 合作伙伴购买 Azure 服务。

浏览你的选项

其他资源

Azure OpenAI 服务

详细了解 Azure OpenAI 服务 特性和功能。

定价计算器

估计每月使用任何 Azure 产品组合应产生的费用。

SLA

查看 Azure OpenAI 服务 的服务级别协议。

文档

查看技术教程、视频和更多 Azure OpenAI 服务 资源。

  • Azure OpenAI 服务根据即用即付和预配吞吐量单位 (PTU) 提供定价。即用即付允许为使用的资源付费,使其能够灵活应对可变的工作负载。PTU 提供可预测的定价模型,可在其中预留和部署特定数量的模型处理容量。此模型非常适合具有一致或可预测使用模式的工作负载,提供稳定性和成本控制。
  • 要详细了解 PTU 和 Azure OpenAI 定价,请阅读 PTU 文档 或 联系我们的销售专家。

与销售专家交谈,演练 Azure 定价情况。了解你的云解决方案的定价。

获取免费云服务和价值 $200 的赠金来探索 Azure 30 天。

添加到估价。 按“V”在计算器上查看
中文(中国)
你的隐私选择选择退出图标 你的隐私选择
消费者健康隐私 与 Microsoft 联系 隐私 管理 Cookie 使用条款 商标 关于我们的广告 京ICP备09042378号-6