GPT-4o mini is a small, cost-optimized large language model released by OpenAI in July 2024 as a lighter counterpart to GPT-4o. It replaced GPT-3.5 Turbo as the default model for free-tier ChatGPT users and as OpenAI's entry-level API offering, at a developer-reported launch price of $0.15 per million input tokens and $0.60 per million output tokens, more than an order of magnitude cheaper than frontier models of the period.1
Design and capabilities
Like its parent model, GPT-4o mini is a natively multimodal Transformer that accepts text and image inputs and produces text outputs; OpenAI stated that support for additional modalities was planned. Its parameter count and training details are undisclosed, though commentators widely assumed some form of distillation or size reduction relative to GPT-4o. The model offers a 128,000-token context window, a maximum of 16,384 output tokens per request, and a knowledge cutoff of October 2023.1 OpenAI reported a score of 82% on MMLU, positioning the model above GPT-3.5 Turbo and competitive with contemporary small models.1 It was also the first OpenAI model to ship with instruction-hierarchy training, a safety method intended to make models prioritize system messages over conflicting user or tool inputs.2
Reception and role
GPT-4o mini arrived during an industry-wide shift toward inexpensive "small frontier" models, competing with Google's Gemini 1.5 Flash and, later in 2024, Anthropic's Claude 3.5 Haiku. Its low price made it a default choice for high-volume tasks such as classification, extraction, and simple chat, and OpenAI opened fine-tuning access for the model shortly after launch.13 Within the GPT family it effectively retired the GPT-3.5 line and complemented larger siblings such as GPT-4 Turbo and GPT-4o, as well as the reasoning-focused o1 series introduced later that year. It was subsequently superseded in OpenAI's small-model lineup by the GPT-4.1 mini tier in 2025 and by the GPT-5-generation mini models.