GPT-4o mini is a compact multimodal model developed by OpenAI, designed to provide high intelligence at a significantly lower cost and latency than larger frontier models. Released as a successor to GPT-3.5 Turbo, it is part of the 'omni' model family and supports text and vision input, with future support for additional modalities like audio and video. It is intended for applications requiring high-volume throughput or real-time responses where cost-efficiency is a primary concern. ## Capabilities and Performance The model features a 128,000 token context window and has a knowledge cutoff of October 2023. It utilizes an improved tokenizer shared with GPT-4o, which enhances the processing efficiency of non-English text. In benchmark evaluations, GPT-4o mini achieved a score of 82.0% on MMLU (Massive Multitask Language Understanding) and has shown proficiency in mathematical reasoning and coding tasks compared to other models in its size class. The model also includes safety mitigations such as the instruction hierarchy technique to improve resistance to prompt injection and jailbreaking.
