o3-pro – OpenAI’s latest reasoning model, supporting access to the full suite of ChatGPT tools

What is o3-pro?

o3-pro is a powerful reasoning model launched by OpenAI, serving as an upgraded version of o3. It excels in handling complex problems and providing precise answers, showing clear advantages in fields such as scientific research, programming, education, and writing. o3-pro supports the full suite of ChatGPT tools, including web search, file analysis, image reasoning, and Python programming, delivering stronger overall execution and integration capabilities. Due to more complex task scheduling and toolchain calls, its response speed is slightly slower than o1-pro, making it better suited for scenarios requiring deep thinking or high accuracy. Expert evaluations show that o3-pro outperforms previous models in clarity of expression, answer completeness, and logical accuracy. o3-pro is available to Pro and Team users, replacing the former o1-pro.

Key Features of o3-pro

• Powerful reasoning capability: Excels at solving complex problems and delivering precise answers, especially in areas needing deep analysis and logical reasoning such as scientific research, programming, education, data analysis, and writing.

• Multi-tool integration: Supports the entire ChatGPT toolset, such as web search, file analysis, image reasoning, Python programming, and personalized memory, resulting in stronger execution and integration.

Performance of o3-pro

• Response speed: Slightly slower than o1-pro due to more complex task scheduling and toolchain invocations.

• Expert evaluations: Experts consistently rate o3-pro higher than the o3 model in expression clarity, answer completeness, instruction-following ability, and logical accuracy.

• Multi-domain optimization: Performs better than previous models like o1-pro and o3 in mathematics, science, and programming.

• Mathematics competitions (AIME 2024): o3-pro achieved a 93% success rate, higher than o3’s 90% and o1-pro’s 86%.

• Doctorate-level science questions (GPQA Diamond): o3-pro scored 84%, outperforming o3 (81%) and o1-pro (79%).

• 4/4 reliability evaluation: Under a strict “4/4 reliability” criterion—where a model must answer correctly in four consecutive attempts to be considered successful—o3-pro substantially outperforms o1-pro and o3, becoming OpenAI’s most powerful model for math, research, and coding.

Limitations of o3-pro

• Does not currently support ephemeral conversations (temporary dialogues).

• Does not support image generation; users should use GPT-4o, o3, or o4-mini for generating images.

• Does not support the Canvas feature; users need other models that support this function.

Pricing for o3-pro

• Input tokens: $20 per million tokens.

• Output tokens: $80 per million tokens.

Project website

https://help.openai.com/en/articles

Application Scenarios for o3-pro

• Data interpretation: Assists users in understanding and analyzing complex datasets, offering suggestions and explanations for data visualization.

• Code optimization: Provides suggestions to improve code performance and readability for developers.

• Personalized learning: Offers customized learning advice and tutoring based on students’ progress and comprehension.

• Business consulting: Provides business advice and solutions to help optimize workflows and improve efficiency.

• Image analysis: Understands and analyzes image content, answers questions related to images, and supports complex visual tasks such as image interpretation and chart analysis.