Following its preview at the December ship-mas event, OpenAI is making the o3-Mini reasoning model available to all users. This model provides current answers and performs exceptionally well on arithmetic and coding problems, but it is blind. It is available to users for free through the ChatGPT platform.
OpenAI is finally making the o3-Mini reasoning model available to both free and premium users after showcasing it during the 12-day Ship-mas event in December. The launch of OpenAI's first-ever "free" reasoning model coincides with the business's growing competition from DeepSeek, a Chinese company whose R1 reasoning model has been publicly accessible since its early January launch.According to OpenAI, the new o3-Mini model has been tuned for activities like science, math, and coding. It surpasses o1-Mini by a wide margin with "more accurate and clearer answers" and performs on par with o1 in its reasoning and evaluation tests.
o3-Mini has also been integrated to offer "up-to-date answers with links to relevant web sources" through ChatGPT's web search technology. Users won't be able to send photographs or ask the model questions, though, because the new model lacks visual capabilities.
What are models of reasoning?
Large language models that have been trained using reinforcement learning to carry out challenging tasks are known as reasoning models. In contrast to standard models such as the GPT-4, reasoning models are trained to generate a lengthy internal chain of thought prior to reacting. This enables them to perform better on tasks like problem solving, coding, scientific reasoning, and multi-step planning, even though it slows down their response time.
Although the performance of OpenAI's new o3-mini model is nearly identical to that of DeepSeek R1, it is twice as expensive at $1.10 for one million input tokens as its Chinese equivalent. Notably, US AI businesses have been under more scrutiny after DeepSeek's V3 and R1 models were recently released.
In addition to upending preconceived ideas about American AI dominance, DeepSeek stated in a study that, in contrast to OpenAI and others, which have spent over $100 billion on comparable models, it only invested $5.6 million and trained its V3 model on older Nividia H800 GPUs.
How is o3-Mini used?
The first reasoning model accessible to free ChatGPT users is o3-Mini. Simply click on the new 'Reason' option beneath the chat bar on the official ChatGPT website or app to access the model. It is reasonable to presume that the consumption restriction for free users would be comparable to the levels of the GPT-4o model, even if OpenAI did not specify it in its blog post. In the meantime, OpenAI has tripled ChatGPT's cap.
Developers can select between three performance levels for o3-Mini based on their unique use cases: low, medium, and high. However, 'unlimited access' to o3-mini and o3-mini high will only be available to ChatGPT Pro customers.
0 Comments