1. News
  2. GPT-4o mini: Accessible AI

OpenAI launches GPT-4o mini: Powerful and Accessible AI | | | | Turtles AI

OpenAI launches GPT-4o mini: Powerful and Accessible AI
A new era for AI: High performance at low costs

Highlights:

  • Reduced costs: GPT-4o mini is 60% cheaper than GPT-3.5 Turbo.
  • High performance: Surpasses other models in academic and coding benchmarks.
  • Versatility: Supports textual and visual inputs and outputs, with future updates for video and audio.
  • Safety: Improved resistance to jailbreak attempts and prompt injections.

 

OpenAI introduces GPT-4o mini: power and accessibility for AI

 

OpenAI has launched GPT-4o mini, a small yet powerful AI model that promises to significantly expand AI applications due to its reduced costs and high performance. This model is designed to be economical and versatile, suitable for a variety of tasks.

 

In the context of rapid technological development, OpenAI has unveiled GPT-4o mini, an AI model distinguished by efficiency and low costs. With an accuracy of 82% on MMLU, GPT-4o mini surpasses GPT-3.5 Turbo and proves superior in academic benchmarks, confirming its validity in text and multimodal reasoning tasks. The cost is significantly reduced to 15 cents per million input tokens and 60 cents per million output tokens, making it about 60% cheaper than GPT-3.5 Turbo.

 

GPT-4o mini is ideal for applications requiring low latencies and operational costs. It can be used for multiple model calls, handling large volumes of context, and rapid customer interactions, such as in customer support chatbots. The model supports textual and visual inputs and outputs, with future goals of including video and audio inputs. The context window is 128K tokens, supporting up to 16K output tokens per request, with knowledge updated to October 2023.

 

In terms of performance, GPT-4o mini surpasses other similar models like Gemini Flash and Claude Haiku in various benchmarks. For example, in MGSM, which measures mathematical reasoning, GPT-4o mini scored 87%, compared to 75.5% for Gemini Flash and 71.7% for Claude Haiku. In the HumanEval test, which assesses coding capabilities, GPT-4o mini achieved 87.2%, compared to 71.5% for Gemini Flash and 75.9% for Claude Haiku.

 

During development, OpenAI partnered with companies like Ramp and Superhuman to test the model in real-world applications, achieving results superior to GPT-3.5 Turbo in structured data extraction and high-quality email response generation. Safety is a priority integrated at every stage of the development process. Using advanced techniques such as reinforcement learning with human feedback, OpenAI improved the reliability and safety of the model’s responses, better resisting jailbreak attempts and prompt injections.

 

GPT-4o mini is now available through the Assistants, Chat Completions, and Batch APIs. Pricing is 15 cents per million input tokens and 60 cents per million output tokens. ChatGPT users, both Free, Plus, and Team, can access GPT-4o mini starting today, while Enterprise users will have access next week. This launch marks an important step toward a future where AI models are integrated into every application and website, making AI more accessible and embedded in daily digital experiences.