AI BriefingGPT-4o mini: The Evolution of Cost-Efficient Intelligence
Key point
OpenAI has unveiled **GPT-4o mini**, a small model that is over 60% cheaper than GPT-3.5 Turbo while offering enhanced performance.
Details
OpenAI has announced GPT-4o mini, its most cost-efficient small model. The model is priced at 15 cents per 1 million input tokens and 60 cents per 1 million output tokens, making it much cheaper than existing frontier models and delivering a cost reduction of over 60% compared to GPT-3.5 Turbo.
With its low cost and low latency, GPT-4o mini is optimized for a variety of applications, including multiple API calls, passing large amounts of context, and real-time customer support chatbots. It currently supports text and vision via the API, with audio and video input and output capabilities to be added in the future.
Key performance metrics are as follows:
- MMLU (reasoning): Scored 82.0%, surpassing Gemini Flash (77.9%) and Claude Haiku (73.8%).
- MGSM (math): Achieved 87.0%, outperforming Gemini Flash (75.5%).
- HumanEval (coding): Reached 87.2%, overwhelming existing small models.
- MMMU (multimodal reasoning): Scored 59.4%.
The model supports a 128K context window and up to 16K output tokens, with knowledge up to October 2023. It also features an improved tokenizer that increases cost efficiency for processing non-English languages.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.