Photo credit: www.engadget.com

OpenAI Launches o3-mini: A Step Forward in Machine Learning

OpenAI has unveiled its latest machine learning model, o3-mini, which became available for users on Friday. In a notable first, the model is accessible to free users of ChatGPT, marking a significant shift in OpenAI’s approach. Users interested in exploring this new functionality can do so by selecting the “Reason” button within the message composer.

OpenAI claims that o3-mini outperforms its predecessor, o1-mini, in both speed and accuracy. During A/B testing, results indicated that o3-mini delivers responses 24 percent faster than o1-mini. Furthermore, under a “medium” reasoning setting, o3-mini approaches the performance levels of the more advanced and costly o1 system, particularly in disciplines like math, coding, and science. In line with OpenAI’s tradition, this model provides users with insights into its reasoning process rather than merely presenting answers. An added feature includes integration with ChatGPT Search, which allows the model to access real-time web information and relevant resources. OpenAI has committed to enhancing search functionality across all its reasoning models in the future.

The introduction of o3-mini plays an essential role in OpenAI’s ongoing mission to innovate while ensuring cost-effectiveness. Company representatives stated, “By optimizing reasoning for STEM domains while keeping costs low, we’re making high-quality AI even more accessible.” The model signifies a continued effort to make AI more affordable, reflecting a remarkable 95% decrease in per-token pricing since the launch of GPT-4, all while maintaining competitive reasoning capabilities. OpenAI is focused on leading advancements in AI that balance intelligence, efficiency, and safety at scale.

With the launch of o3-mini, it takes the place of o1-mini in the available model options. Moreover, OpenAI is increasing the limit on message usage for Plus and Team ChatGPT users, tripling the allowance from 50 messages daily with o1-mini to 150 messages daily with o3-mini. The new Pro tier, priced at $200 per month, grants users unrestricted access to the latest model.

When OpenAI initially introduced o3 and o3-mini late last year, CEO Sam Altman suggested that o3-mini would be available “around the end of January.” A more specific timeline was provided on January 17 when Altman posted on social media about plans to release the model within weeks.

The urgency surrounding the launch is palpable, especially considering the competitive landscape. On January 20, while Altman was present at Donald Trump’s inauguration, DeepSeek, a Chinese AI company, released its own R1 chain-of-thought model. By January 27, DeepSeek’s chatbot achieved significant popularity, surpassing ChatGPT as the most downloaded free application on the US App Store. This rapid rise had a substantial impact, erasing $1 trillion from stock market valuations, likely catching OpenAI off guard.

Following the rise of DeepSeek, OpenAI announced its collaboration with Microsoft to trace two specific accounts that allegedly may have utilized its models for distillation — a process where knowledge is transferred from a larger, more complex AI system to a more streamlined version. While distillation itself is widely accepted in the industry, DeepSeek employed this method to enhance its R1 model while training smaller algorithms. OpenAI’s terms permit such use, provided that outputs from its AI are not exploited to train competing frameworks.

Although OpenAI has not directly accused DeepSeek, they have acknowledged an ongoing concern. An OpenAI spokesperson commented on the situation, noting, “We know [China]-based companies — and others — are constantly trying to distill the models of leading US AI companies.” Reflecting on the matter, David Sacks, a notable advisor in AI, asserted that there is “substantial evidence” suggesting DeepSeek had indeed distilled OpenAI’s models.

Source
www.engadget.com