Transforming the Title into a Unique Variation with GPT-4 Mini

OpenAI is further democratizing access to cutting-edge AI technology with its recent introduction of GPT-4o mini, which is now available through its API and being incorporated into ChatGPT as of today.

The latest compact model from OpenAI, GPT-4o mini, combines advanced intelligence with budget-friendly accessibility. It outshines its predecessor, GPT-3.5 Turbo, in terms of both intelligence and affordability, marking a significant advancement in AI technology.

With the introduction of GPT-4o mini, OpenAI solidifies its commitment to broadening the reach of artificial intelligence. This highly efficient and economical small model paves the way for a broader spectrum of AI-driven applications by making advanced computational intelligence more accessible. GPT-4o mini has achieved a remarkable score of 82% on the MMLU benchmark, surpassing GPT-41 in chat preferences on the LMSYS leaderboard. Its cost-effective pricing at 15 cents per million input tokens and 60 cents per million output tokens positions it substantially ahead of earlier models, offering over 60% savings compared to GPT-3.5 Turbo.

Thanks to its affordability and lower latency, GPT-4o mini facilitates diverse applications, including complex operations like chaining or running multiple APIs concurrently, managing extensive data inputs, or powering live customer interaction tools such as support chatbots.

The API of GPT-4o mini

At present, GPT-4o mini supports both text and vision functionalities within the API, with future expansions planned to embrace text, image, video, and audio inputs and outputs. The model provides a sizable context window of 128K tokens and incorporates knowledge updated to October 2023. The enhanced tokenizer, shared with GPT-4o, offers improved efficiency in managing texts in languages other than English.

As a model of modest size yet superior in textual comprehension and multimodal reasoning, GPT-4o mini stands head and shoulders above GPT-3.5 Turbo and similar small-sized models in academic evaluations. It supports a broad spectrum of languages as GPT-4o and is adept at function calling, which empowers developers to craft apps capable of data retrieval or external system interactions, also showing advancements in managing long-context information over GPT-3.5 Turbo.

Assessment of GPT-4o mini

A comprehensive review of GPT-4o mini across key benchmarks reveals:

• Cognitive Reasoning: In tasks requiring cognitive reasoning across text and visuals, GPT-4o mini leads, with an 82.0% success rate on the MMLU benchmark, surpassing models like Gemini Flash and Claude Haiku.
• Math and Computational Tasks: GPT-4o mini showcases exceptional abilities in mathematics and coding, outperforming the competition significantly in both fields.
• Multimodal Understanding: This model also exhibits robust capabilities in understanding and analyzing multimodal content, scoring higher than its competitors in similar assessments.

In their model development phase for GPT-4o mini, OpenAI collaborated with trusted partners including Ramp and Superhuman, discovering its superiority in tasks like data extraction from receipts and crafting quality email responses based on conversation histories, outshining GPT-3.5 Turbo.

Integrated Safety Protocols

Right from the start, OpenAI integrates safety measures into their models, refining these safeguards at every stage of development. During the pre-training phase, they exclude undesirable data like hate speech and adult content. Post-training, adjustments are made to align the model more closely with desired behavioral standards using techniques like reinforcement learning with human feedback (RLHF), enhancing the model’s response accuracy and reliability.

Encapsulating the same safety features as GPT-4o, GPT-4o mini underwent thorough automated and human evaluations under OpenAI’s Preparedness Framework. Insights from over 70 external specialists have informed the enhancement of safety features in GPT-4o mini, making it the first model to apply an instruction hierarchy method for better safety in widespread application use.

Continuous observation and enhancement of GPT-4o mini’s safety measures remain a priority for OpenAI as they navigate new risk landscapes.

How to Access and Pricing Plans

Available for integration in various applications through the Assistants API, Chat Completions API, and Batch API, GPT-4o mini is priced competitively for developers. The access extends across different tiers in ChatGPT, affirming OpenAI’s mission towards universal AI accessibility.

Looking Ahead

OpenAI continues to observe significant advancements in AI, achieving dramatic reductions in costs. With the GPT-4o mini, OpenAI sets the stage for the next wave of efficient and cost-effective AI application development, signaling a future where AI technologies become integral to online experiences. OpenAI remains at the forefront of this innovative journey, eager to unlock new possibilities.

Related articles: