OpenAI has launched o3-Mini as a faster and more affordable reasoning model geared toward science, math, and coding challenges.
This move arrives while DeepSeek positions itself as the most downloaded chatbot in Apple’s App Store and capitalizes on an open-source R1 system that undercuts OpenAI on price. The intensity in the AI realm is heightened further by revelations of a ChatGPT security flaw called Time Bandit.
o3-Mini’s Pricing and Benchmarks
Microsoft recently made OpenAI’s o1 model free for all Copilot users and removed the requirement for a Copilot Pro plan, complicating OpenAI’s subscription offerings.
Shortly after, OpenAI introduced o3-Mini at 1.10 dollars per million input tokens and 4.40 dollars per million output tokens, a cut of 63 percent compared to the older o1-Mini.
OpenAI o3-mini is now available in ChatGPT and the API.
— OpenAI (@OpenAI) January 31, 2025
Pro users will have unlimited access to o3-mini and Plus & Team users will have triple the rate limits (vs o1-mini).
Free users can try o3-mini in ChatGPT by selecting the Reason button under the message composer.
The new model features a “reasoning effort” setting so users can pick between speedy answers or more thorough responses. User Access to o3-Mini Free ChatGPT users can select “Reason” to try o3-Mini’s medium effort setting, while Plus and Team subscribers get extra daily queries.
Education and enterprise clients will follow soon, with Pro subscribers unlocking the highest reasoning tier.
Free users can now try OpenAI o3-mini in ChatGPT by selecting the Reason button under the message composer. https://t.co/Ws5rXqnKep pic.twitter.com/vT2snEUbzH
— OpenAI (@OpenAI) January 31, 2025
o3-Mini delivers faster responses and better STEM outcomes than its predecessor. The model reportedly achieves fewer major mistakes on coding tasks and math problems, including competition sets such as AIME 2024.




In describing o3-Mini’s scope, OpenAI wrote, “While o1 remains our broader general-knowledge reasoning model, o3-mini provides a specialized alternative for technical domains requiring precision and speed.”
They also explained that o3-Mini can sometimes surpass o1 when set to high reasoning effort, though DeepSeek’s R1 continues to outperform o3-Mini on some advanced tests if given low reasoning time.
In ChatGPT, OpenAI sets o3-Mini’s medium effort by default, which it calls “a balanced trade-off between speed and accuracy.” The same announcement claims that tests on real-world queries, including physics and software engineering prompts, demonstrated o3-Mini’s potential for fewer factual errors.
The DeepSeek Challenge
Competition from DeepSeek’s Open-Source Model DeepSeek continues to expand its foothold despite reports that the chatbot fails 83 percent of accuracy tests.
This discrepancy did not stop Microsoft from adding DeepSeek R1 to Azure AI Foundry even as OpenAI launched an internal review to determine whether DeepSeek may have illegally accessed OpenAI’s training data during the development of its models. Italy has also opened an investigation into DeepSeek over possible GDPR violations, and the U.S. Navy decided to ban DeepSeek’s AI models for security reasons.
Kevin Weil, OpenAI chief product officer, said in a CNBC interview yesterday, “Our mission is to make sure the U.S. wins this race.”
ChatGPT’s Time Bandit Exploit
While o3-Mini aims to keep OpenAI on course, researchers discovered the Time Bandit exploit in ChatGPT, wherein timeline confusion can trick the model into divulging banned information.
Cybersecurity expert David Kuszmar commented the issue with, “Horror. Dismay. Disbelief. For weeks, it felt like I was physically being crushed to death.”
OpenAI acknowledged the vulnerability in a formal CERT note and commented, “We appreciate the researcher for disclosing their findings. We are continuously working to make our models safer and more robust against exploits, including jailbreaks, while also maintaining the models’ usefulness and task performance.”
Partial mitigations have been deployed, although the Time Bandit flaw can still be reproduced in some scenarios.
.