OpenAI has announced the release of GPT-4o mini, a highly cost-efficient small model designed to make artificial intelligence (AI) more accessible and affordable for a wide range of applications.

This model, which excels in both textual intelligence and multimodal reasoning, is priced at15 cents per million input tokens and 60 cents per million output tokens. This marks a dramatic reduction in cost, making it an order of magnitude more affordable than previous frontier models and over 60% cheaper than GPT-3.5 Turbo.

Image Source: Envato

Capabilities and Performance

Despite its smaller size, GPT-4o mini boasts impressive capabilities:

  • Text and Vision Support: Currently, GPT-4o mini supports text and vision in the API, with plans to include image, video, and audio inputs and outputs in the future.
  • Extended Context Window: The model features a context window of 128K tokens and can support up to 16K output tokens per request.
  • Multilingual Proficiency: With an enhanced tokenizer, GPT-4o mini handles non-English text efficiently, broadening its applicability across different languages.

GPT-4o mini outperforms other small models on academic benchmarks, scoring 82% on the MMLU benchmark, surpassing GPT-3.5 Turbo, Gemini Flash, and Claude Haiku. In mathematical reasoning and coding tasks, it leads with scores of 87.0% on MGSM and 87.2% on HumanEval.

Applications and Use Cases

The affordability and efficiency of GPT-4o mini make it ideal for a variety of applications:

  • High-Volume Contextual Analysis: Suitable for processing large volumes of context, such as full code bases or conversation histories.
  • Real-Time Interactions: Effective for real-time customer support chatbots and other interactive applications.
  • Chained and Parallel Model Calls: Facilitates tasks requiring multiple API calls, enhancing workflow automation and efficiency.

Safety and Reliability

Safety is a cornerstone of GPT-4o mini’s development. OpenAI has integrated comprehensive safety measures, including:

  • Pre-Training Filters: Excluding harmful content such as hate speech and adult content from training data.
  • Post-Training Alignments: Using reinforcement learning with human feedback (RLHF) to align the model’s behavior with safety policies.
  • Instruction Hierarchy Method: This innovative technique improves resistance to jailbreaks and prompt injections, making GPT-4o mini’s responses more reliable.

GPT-4o mini is available through the Assistants API, Chat Completions API, and Batch API. Starting today, Free, Plus, and Team users of ChatGPT can access GPT-4o mini, with Enterprise access following next week. OpenAI also plans to introduce fine-tuning capabilities for GPT-4o mini soon.

Read AI Briefings