Home » News » OpenAI Launches o3-Mini Models: Free AI Model to Take on DeepSeek

OpenAI Launches o3-Mini Models: Free AI Model to Take on DeepSeek

by Ravi Teja KNTS
0 comment

DeepSeek R1 shook the AI industry last week by offering its o1-level reasoning model for free and open-sourced it in the process. OpenAI has responded by unveiling two new reasoning models—o3-Mini and o3-Mini-High.

In what appears to be a direct response to DeepSeek, OpenAI is making the o3-Mini model available for free. This is the first time free users can access a reasoning model from OpenAI. But what sets these models apart, and what do they mean for you? Let’s break it down.

What are o3-Mini and o3-Mini-High?

OpenAI first teased the o3 family of models in December 2024 during its 12 Days of OpenAI campaign. At the time, they revealed that o3-Mini would come in three reasoning tiers—low, medium, and high.

The model was launched today as o3-Mini falls under the medium reasoning tier and is available to all ChatGPT users, including free users. Meanwhile, o3-Mini-High, which offers higher reasoning capabilities, is exclusive to Pro, Plus, and Teams users.

o3-Mini (Medium Effort Reasoning Model)

OpenAI’s o3-Mini is a cost-effective reasoning model optimized for STEM tasks, including math, coding, and science. According to OpenAI, A/B testing showed that o3-Mini delivers responses 24% faster than o1-Mini, with an average response time of 7.7 seconds compared to 10.16 seconds.

In terms of performance, o3-Mini not only surpasses o1-Mini but also reaches o1-level reasoning, particularly excelling in coding and PhD-level science questions. On Codeforces benchmarks, o3-Mini scored 2036, significantly outperforming the 1891 score of the o1 model. Simply put, o3-Mini is faster than o1-Mini but performs better than o1.

For access, Plus and Teams subscribers get 150 messages per day, while Pro users enjoy unlimited messages. Free users, however, are subject to rate limits similar to GPT-4o’s—roughly 10 to 16 messages per three to five hours. OpenAI hasn’t disclosed exact limits, so this remains an estimate.

o3-Mini-High: The Best Publicly Available Reasoning Model

While o3-Mini matches o1-level performance in benchmarks, o3-Mini-High goes a step further, surpassing o1 model and therefore the DeepSeek R1. Like its counterpart, this model also significantly improves coding, achieving a Codeforces benchmark score of 2130, compared to 1891 for o1, similar to DeepSeek R1.

Currently, o3-Mini-High seems to be the best publicly available reasoning model but we will pitch it against other reasoning models soon. However, access is limited to Plus, Teams, and Pro users. Pro users have unlimited access, while Plus and Teams users have a 150-message daily limit.

OpenAI has also made o3-Mini available through its API, offering cheaper and faster AI services. Compared to o1-Mini, the new model is 63% more cost-effective.

Also Read:

How to Access o3-Mini in ChatGPT

Free users can access the o3-Mini model by simply enabling the Reason option in the chatbox.

Paid users (ChatGPT Plus, Pro, and Teams) can select either o3-Mini or o3-Mini-High from the model picker, alongside other available models.

In our experience, o3-Mini performs remarkably well. However, we’ll have to wait for the full release of the o3 model to see if it delivers on OpenAI’s promise of AGI-level AI.

You may also like