Key points:

  • OpenAI introduces the GPT-4o mini, an affordable option for companies on a budget.
  • OpenAI, a leader in AI software backed by Microsoft, is taking steps to reduce the cost and speed up application development.
  • The company is also developing its own AI chip, in talks with Broadcom.

On Thursday 18, OpenAI, the developer of the popular chatbot ChatGPT, announced the release of GPT-4o mini. This new model is a compact and cost-effective version of AI designed to make OpenAI technologies more accessible and energy efficient.

Cost optimization

OpenAI, a leader in AI software backed by Microsoft, is taking steps to reduce the cost and speed up application development based on its developer model. It comes amid efforts by deep-pocketed rivals such as Meta and Google to grab a larger share of the market.

OpenAI’s new GPT-4o mini model offers significant cost reductions over its predecessor: 15 cents per million input tokens and 60 cents per million output tokens, which is 60% cheaper than the GPT-3.5 Turbo.

In addition to being cost-effective, GPT-4o mini demonstrates superiority over GPT-4 in chat settings, reaching 82% in the MMLU (Mass Multitask Language Understanding) test. MMLU assesses the textual intelligence and reasoning of language models, and a higher score indicates a better ability to understand and use language across domains, increasing its application to real-world problems.

The GPT-4o mini beats its competitors: Google’s Gemini Flash (77.9%) and Anthropic’s Claude Haiku (73.8%).

The small size of the model makes it less demanding on computing resources, which positions the GPT-4o mini as an affordable option for companies on a limited budget who want to implement generative AI in their work.

OpenAI is in talks with Broadcom to develop a new AI chip

OpenAI is also in talks with chip developers including Broadcom to create a custom AI chip, The Information newspaper reported.

The motivation for this move was the shortage of powerful and expensive GPUs that are used to develop and train artificial intelligence models such as ChatGPT, GPT-4 and DALL-E3.

Microsoft-backed OpenAI is recruiting former Google engineers who once created the Tensor Processor, the search giant’s AI chip. According to the source, the goal is to develop a server chip optimized for artificial intelligence tasks.

Bloomberg News previously reported on OpenAI CEO Sam Altman’s plans to attract billions of dollars in investment to create a network of semiconductor factories. Intel, Taiwan Semiconductor Manufacturing Co and Samsung Electronics were named as potential partners in this project.