Hajra Naz
On Thursday, OpenAI unveiled GPT-4o mini, its latest compact AI model aimed at developers and consumers alike. This smaller and swifter model is set to debut on the ChatGPT web and mobile app today, with enterprise access scheduled for next week.
Performance and Use Cases
OpenAI asserts that GPT-4o mini surpasses leading small AI models in tasks involving text and vision reasoning. With improvements in efficiency and speed, such models are increasingly favoured by developers for their cost-effectiveness in high-volume and repetitive tasks.
Benchmarks and Comparison
Replacing GPT-3.5 Turbo as OpenAI’s smallest model, GPT-4o mini boasts significant benchmark achievements. It scores 82% on the MMLU reasoning benchmark, outstripping competitors like Gemini 1.5 Flash and Claude 3 Haiku. Moreover, it achieves an 87% score on MGSM for math reasoning, demonstrating superior performance over previous models.
Affordability and Accessibility
OpenAI highlights the affordability of GPT-4o mini, priced over 60% lower than its predecessor. Initially supporting text and vision capabilities, the model aims to expand to video and audio functionalities in the future.
Future Prospects
Olivier Godement, OpenAI’s head of Product API, emphasized affordability’s importance in democratising global AI access. He views GPT-4o mini as a significant advancement in this direction.
Technical Specifications
Priced at 15 cents per million input tokens and 60 cents per million output tokens, GPT-4o mini features a context window of 128,000 tokens and a knowledge cutoff of October 2023. Although exact dimensions were not disclosed, it is comparable to other small AI models like Llama 3 8b and Gemini 1.5 Flash, yet surpasses them in speed, efficiency, and intelligence according to preliminary testing.
Early Validation
Early tests from LMSYS.org confirm GPT-4o mini’s superior speed, averaging 202 tokens per second, more than double that of GPT-4o and GPT-3.5 Turbo. This speed makes it particularly suitable for applications requiring rapid responses and extensive user interaction.
By leveraging these advancements, OpenAI aims to redefine the landscape of accessible AI solutions, making powerful models like GPT-4o mini more practical and pervasive across various industries and consumer applications.
OpenAI Introduces New Tools for ChatGPT Enterprise
On Thursday, OpenAI launched new tools for enterprise customers. They introduced the Enterprise Compliance API in a blog post. This API is for businesses in regulated sectors like finance, healthcare, legal services, and government. Its goal is to help these businesses meet strict logging and audit requirements.
Features and Functionality
OpenAI’s Enterprise Compliance API empowers administrators to conduct audits and manage ChatGPT Enterprise data effectively. It provides comprehensive records of timestamped interactions, encompassing conversations, uploaded files, workspace users, and more.
Enhanced Administrative Control
In addition to compliance features, OpenAI is enhancing administrative oversight for workspace-specific GPTs. Previously, administrators could only fully permit or block GPT actions within their workspace. Now, workspace owners can establish approved domains with which GPTs can interact, offering more granular control tailored to specific business needs.
These updates underscore OpenAI’s commitment to providing robust tools that meet the rigorous regulatory demands of enterprise environments while enabling tailored and secure interactions with ChatGPT within organizational settings.
