OpenAI has unveiled a smaller and cost-effective model GPT-4o Mini. The company claims that it is cheaper and smaller than GPT-3.5. GPT-4o is priced at 15 cents per million input tokens and 60 cents per million output tokens.
GPT-4o mini can enable a broad range of tasks and scores 82% on the MMLU benchmark, On chat preferences in the LMSYS leaderboard, it can even outperform many other models including GPT-4. So, it is an ideal choice for developers and businesses who want to leverage the power of AI without spending much.
It supports text and vision in the API. The company also has plans for image, video, and audio inputs and outputs, so GPT-4o mini can meet the diverse needs of modern applications.
GPT-4o Mini surpasses other small models in several key areas:
- Reasoning Tasks: Scores 82.0% on MMLU, outperforming Gemini Flash and Claude Haiku.
- Math and Coding Proficiency: Achieves 87.0% on MGSM for math reasoning and 87.2% on HumanEval for coding performance.
- Multimodal Reasoning: Scores 59.4% on MMMU, demonstrating its strong capability in handling both text and vision.
OpenAI has collaborated with trusted partners like Ramp and Superhuman to develop GPT-4o Mini. The company has also incorporated several safety measures from pre-training to post-training to filter out undesirable content.
GPT-4o Mini is available starting today as a text and vision model through the Assistants API. Options to fine-tune will be available in the coming days. ChatGPT users can access GPT-4o Mini immediately and Enterprise users will gain access next week.
