EN
OpenAI has recently introduced three new models within ChatGPT: 03, 04 mini, and 04 mini high. These models are designed to enhance reasoning capabilities, allowing them to think critically before generating responses. The 03 model is set to replace the older 01 model, while the pro plan, priced at $200 per month, will also phase out the 01 pro mode, marking it as a legacy reasoning model.
The latest models from OpenAI are the most advanced to date, showcasing unique abilities that can be explored through various benchmarks. For instance, 03 outperforms 01, and 04 mini is currently the top model available. Notably, 04 mini high stands out as the best option for users. In competitive benchmarks, 04 mini scores an impressive 2719, placing it among the top 200 coders globally, while also achieving over 90% in other assessments.
One of the standout features of these new models is their multimodal reasoning ability, which allows them to solve visual problems. For example, 04 mini scores 81% in this area, while 03 scores slightly higher at 82%. Depending on the specific task, 03 may outperform 04 mini in multimodal scenarios. These models are currently accessible in the pro, teams, and plus plans.
To illustrate the capabilities of 04 mini high, a visual reasoning example was conducted. By uploading an image of a ship, the model successfully identified the ship's name and its next docking location using AIS data. This model autonomously performed a web search to gather the necessary information, showcasing its advanced functionality.
The new memory feature allows the models to reference previous conversations, enhancing their ability to provide personalized responses. For instance, when prompted about recent news, the model utilized its memory and reasoning capabilities to deliver relevant information tailored to the user's interests.
The models also excel in making predictions based on current data. For example, when asked to predict tariff levels between the US and China in June 2025, both 03 and 04 mini high provided consistent estimates after analyzing multiple sources. This demonstrates their ability to synthesize information and deliver informed predictions.
The coding capabilities of the new models were tested with a request for a Galaga-style game. The models generated the necessary code efficiently, showcasing their ability to handle programming tasks. The output was well-structured and functional, indicating the models' proficiency in coding.
In a mathematical problem-solving scenario, 04 mini provided two possible combinations of animals purchased within a specified budget. In contrast, the older 01 Pro model took significantly longer to arrive at the same conclusion, highlighting the efficiency of the newer models.
The models also demonstrate strong estimation skills. For example, when estimating the number of piano tuners in New York City, the model provided a logical answer based on population assumptions. This capability is particularly useful for users seeking quick, data-driven insights.
The interface of ChatGPT has undergone significant changes, making it more user-friendly. The new reasoning models are designed for specific tasks, with 04 mini high being the preferred choice for coding and visual analysis. Users are encouraged to utilize the appropriate model based on their needs to maximize efficiency.
For those new to ChatGPT or seeking to enhance their skills, a beginner prompting course has been introduced. This course covers the basics and is available for a free trial. Additionally, a variety of other courses are offered, catering to both beginners and advanced users, ensuring that everyone can find valuable resources to improve their understanding of ChatGPT.
Q: What are the new models introduced by OpenAI?
A: OpenAI has introduced three new models: 03, 04 mini, and 04 mini high, designed to enhance reasoning capabilities.
Q: How do the new models compare to the older models?
A: The 03 model is set to replace the older 01 model, and the pro plan will phase out the 01 pro mode, marking it as a legacy reasoning model.
Q: Which new model is currently the top performer?
A: The 04 mini is currently the top model available, with 04 mini high being the best option for users.
Q: What are the performance benchmarks for the new models?
A: In competitive benchmarks, 04 mini scores 2719, placing it among the top 200 coders globally, and achieves over 90% in other assessments.
Q: What is multimodal reasoning?
A: Multimodal reasoning allows the models to solve visual problems, with 04 mini scoring 81% and 03 scoring 82% in this area.
Q: Can you provide an example of visual reasoning capabilities?
A: In a demonstration, 04 mini high identified a ship's name and its next docking location by autonomously performing a web search.
Q: How does the memory feature enhance user interaction?
A: The memory feature allows models to reference previous conversations, providing personalized responses based on user interests.
Q: What kind of predictions can the models make?
A: The models can make predictions based on current data, such as estimating tariff levels between the US and China in June 2025.
Q: How proficient are the new models in coding?
A: The models efficiently generated code for a Galaga-style game, showcasing their proficiency in handling programming tasks.
Q: How do the new models perform in mathematical problem-solving?
A: 04 mini provided solutions to mathematical problems more efficiently than the older 01 Pro model.
Q: Can the models estimate real-world scenarios?
A: Yes, the models demonstrate strong estimation skills, such as logically estimating the number of piano tuners in New York City.
Q: What updates have been made to the ChatGPT interface?
A: The interface has become more user-friendly, with specific models designed for tasks, encouraging users to choose the appropriate model.
Q: Are there educational resources available for new users?
A: Yes, a beginner prompting course is available for free trial, along with various other courses for both beginners and advanced users.