Revolutionising AI on a Budget:
Meet GPT-4o Mini 

OpenAI is pushing the boundaries of accessible AI with the introduction of GPT-4o Mini, their most cost-efficient small model to date. Here are my first impressions of this exciting new development. 

Affordable Excellence 

GPT-4o Mini is priced at just 15 cents per million input tokens and 60 cents per million output tokens, making it incredibly budget-friendly. Despite the low cost, it scores an impressive 82% on the MMLU benchmark, outperforming GPT-3.5 Turbo in chat preferences on the LMSYS leaderboard. 

Versatile Applications 

This model’s affordability and low latency make it ideal for applications requiring multiple model calls or large volumes of context, such as customer support chatbots. GPT-4o Mini can currently process text and vision; picture, video, and audio inputs and outputs are planned for the future. It offers up to 16K output tokens per request and a context window of 128K tokens. 

Benchmark Performance 

The GPT-4o Mini performs exceptionally well on a number of academic benchmarks, such as arithmetic, coding, multimodal thinking, and textual intelligence. Here’s a quick look at how it stacks up: 

Built-in Safety 

GPT-4o Mini includes robust safety features, reinforced by extensive evaluations and new techniques like OpenAI’s instruction hierarchy method. This enhances its ability to resist jailbreaks and prompt injections, ensuring safer and more reliable responses. 

Availability 

The Assistants API, Chat Completions API, and Batch API are now available in GPT-4o Mini. With enterprise access beginning next week, it replaces GPT-3.5 in ChatGPT for users who are Free, Plus, and Team members. 

Comparing GPT-4o Mini to GPT-3.5 Turbo 

Advantages

• Cost Efficiency: Significantly cheaper, enabling broader accessibility. 

• Enhanced Performance: Outperforms GPT-3.5 Turbo on multiple benchmarks.

• Extended Context Window: Handles larger volumes of context.

• Versatility: Supports text and vision with plans for more input types.

Disadvantages

• Newer Model: May need further real-world testing for reliability.
• Feature Set: Currently supports fewer input and output types compared to GPT-3.5 Turbo’s established ecosystem.

Final Thoughts 

GPT-4o Mini represents a major leap in making advanced AI both affordable and powerful. Developers and enterprises alike can explore new avenues with its remarkable performance and affordable price. The future of AI is here, and it’s more accessible than ever. Stay tuned for more exciting updates on GPT-4o Mini!