OpenAI is back in the spotlight with the release of two new reasoning models, “o3” and “o4-mini,” featuring unique capabilities including “visual thinking,” code execution, and multimodal analysis.
Article contents:
- Introduction
- Features of the new models
- Competition in the reasoning model market
- Implications for the Arab world
- Conclusion
- Frequently asked questions
Introduction
As competition heats up in the field of artificial intelligence, particularly in the reasoning models sector, OpenAI has rolled out two new versions: “o3” and “o4-mini.” “o3” stands as the most advanced model yet in OpenAI’s model series, with the company announcing that it outperforms all previous models in mathematics, programming, scientific reasoning, and image understanding. Meanwhile, “o4-mini” offers a notable balance between cost, speed, and performance, making it an attractive choice for developers looking to integrate artificial intelligence into their applications without massive budgets.
Artificial intelligence thinks in two ways: explore the secrets of Claude 3.7
Features of the new models
These two models distinguish themselves from their predecessors with the ability to “think with images,” meaning they can read and analyze uploaded images during the pre-response phase. For example, a user can upload an image of a vague drawing or a screenshot, and the model will process the information to reach conclusions or clarifications, such as fixing a code snippet shown in a screenshot. Furthermore, the model can execute Python commands directly within the browser via the “Canvas” feature integrated with ChatGPT, and it can also conduct web searches when up-to-date information is needed.
OpenAI did not stop there, announcing a sub-version of “o4-mini” called “o4-mini-high,” which takes longer to generate answers but increases reliability, especially in sensitive matters requiring extra scrutiny. “o3,” “o4-mini,” and its sub-version are slated to be available to subscribers of ChatGPT Pro, Plus, and Team plans, as well as through APIs that allow developers to call the model at a specific cost per million input or output tokens.
Competition in the reasoning model market
This move highlights OpenAI’s persistent drive to keep pace with major competitors such as Google, Meta, Anthropic, and xAI, who are working on reasoning models that match or outperform OpenAI’s offerings. Although “o3” was nearly left out of ChatGPT due to the company’s desire to integrate it with a broader model, competitive pressures drove its release. The company notes that “o3” achieved 69.1% on the SWE-bench benchmark for verified code, outperforming “o4-mini,” which scored 68.1%, by a small margin.
Notably, OpenAI CEO Sam Altman pointed out that “o3” and “o4-mini” could be the last standalone reasoning models to be integrated into ChatGPT before the launch of GPT-5, the future model in which the company intends to fully integrate reasoning capabilities.
Implications for the Arab world
For users in the Arab world, this development represents a greater opportunity to leverage artificial intelligence in their projects, whether in developing applications with Arabic interfaces or in executing academic and innovative projects that require advanced data or text analysis. However, the challenge of training these models on Arabic-language data remains, particularly in specialized fields.
Ernie 4.5: How Baidu is competing with China’s most advanced AI models
Conclusion
Ultimately, the launch of “o3” and “o4-mini” reveals a daily accelerating pace in the world of artificial intelligence, as companies race to deliver the best models in reasoning and deep learning. Awaiting what OpenAI will present in its future GPT-5 release, users and developers remain the primary beneficiaries of the evolution of these models, provided attention is paid to the ethical and regulatory aspects accompanying this technological race.
Frequently asked questions
1. What is the difference between the “o3 model” and “o4-mini”?
“o3” offers higher performance in programming and reasoning benchmarks, while “o4-mini” provides a better balance between price and speed.
2. How does the “visual thinking” feature work?
The model can analyze images before providing answers, helping it understand visual context and perform tasks like code debugging.
3. Are these models available for public use?
Yes, they can be accessed through paid ChatGPT subscriptions and APIs provided by OpenAI.
4. What is the cost of use for developers?
The price for inputting one million tokens is $10 for the “o3” model and $1.10 for “o4-mini,” with different pricing for output tokens.
5. Is there a plan to release “o3-pro”?
Yes, OpenAI announced its intention to release an “o3-pro” version with higher computing resources dedicated to ChatGPT Pro subscribers.
6. What is the future of these models after GPT-5?
Speculation suggests that GPT-5 will combine reasoning and language capabilities into a single model, which could make “o3” and “o4-mini” the final standalone models.
7. How does the Arab user benefit?
These models can be employed to build Arabic text and image analysis applications, though this requires the availability of high-quality Arabic data.
8. Are there ethical challenges?
Of course, indiscriminate use of reasoning models may raise issues related to privacy and over-reliance on smart technologies.
