GPT-4o Mini (Azure)

GPT-4o mini became available on Microsoft Azure OpenAI Service on July 18, 2024. The model offered text processing with higher speed, lower cost, and stronger multilingual support, while image, audio, and video capabilities were described as forthcoming. The deployment offered a 128K-token context window, fine-tuning, and an 82% MMLU score compared with 70% for GPT-3.5 Turbo. Azure listed pricing of $0.15 per million input tokens and $0.60 per million output tokens under global pay-as-you-go.

Microsoft announced on July 18, 2024 that OpenAI's GPT-4o mini was available on Azure AI, with text processing supported at launch and image, audio, and video planned for later. The model could be tried at no cost in the Azure OpenAI Studio Playground. Alongside it, Microsoft added safety-by-default features for GPT-4o mini, expanded data residency, worldwide pay-as-you-go availability, and performance upgrades to Azure OpenAI Service. Azure's post described GPT-4o mini as significantly smarter than GPT-3.5 Turbo, scoring 82% on MMLU versus 70%, while being more than 60% cheaper. The model expanded the context window to 128K tokens and carried over GPT-4o's improved multilingual capabilities. Microsoft said it saw strong results testing GPT-4o mini with GitHub Copilot for low-latency code completion.

Artificial Analysis lists GPT-4o mini as an OpenAI proprietary model released in July 2024 that is now deprecated, with benchmarking reduced to every 18 hours, and recommends considering the newer GPT-5 mini release instead. The page tracks it as a non-reasoning model accepting text and image input and producing text output, with a 128K context window and an October 2023 knowledge cutoff. Reported performance and pricing include a 1.68-second time to first answer token and 59 tokens per second output speed, at $0.15 per million input tokens and $0.60 per million output tokens with a 50% cache discount. Because the model is flagged as deprecated, these figures are best treated as lifecycle context rather than current recommended deployment guidance.

Microsoft announced that GPT-4o mini became available through the Azure OpenAI API with vision support for Global and East US regional standard deployments. The update added image input and text output capabilities and introduced both global and regional pricing options for developers using the model. The announcement also covered fine-tuning support for text workloads. GPT-4o mini was identified as OpenAI’s fast model, and Azure customers could access it through Azure OpenAI Studio, extending the model beyond the Playground to production API integrations.

Microsoft's Learn documentation provides a walkthrough for fine-tuning the gpt-4o-mini-2024-07-18 deployment on Azure OpenAI, covering creation of sample training and validation datasets, environment variables for the resource endpoint and API key, uploading those files, creating a fine-tuning job, and deploying the resulting custom model. The tutorial lists prerequisites including an Azure subscription, Python 3.8 or later with json, requests, tiktoken, openai and numpy libraries, Jupyter notebooks, and a resource in a region where gpt-4o-mini-2024-07-18 fine-tuning is available. Foundry User role is required, and its test run billed 48,000 tokens for 4,800 training tokens across 10 epochs.

Microsoft's Azure AI Foundry Blog announced on July 31, 2024 that GPT-4o mini moved from the Azure OpenAI Studio Playground into general API availability, adding vision input alongside text output for Global Standard and East US Regional Standard deployments. The post notes OpenAI created the model and Azure hosts it. Developers could then process visual data and documents to extract insights and generate text responses, with global and regional price tiers introduced because regional operating costs and compliance and residency requirements differ. Global Standard was positioned as the lowest-price, highest-throughput starting point, with more regions slated to follow.

Microsoft Foundry's model catalog lists gpt-4o-mini as a Direct from Azure model, version 2024-07-18, sold and managed through Azure under a single license with unified billing and PTU portability. The listing frames it as an affordable, efficient option for diverse text and image tasks, managed on Azure's enterprise-grade infrastructure. The catalog reports vendor benchmark figures for gpt-4o-mini, including MMLU 82.0% versus Gemini Flash 77.9% and Claude Haiku 73.8%, MGSM 87.0%, HumanEval 87.2%, and MMMU 59.4%. It also notes strong function-calling performance for building apps that act on external systems, improved long-context behavior over GPT-3.5 Turbo, and the same language coverage as GPT-4o.