OpenAI o3 and o4-mini

OpenAI has brought out two new AI reasoning models, o3 and o4-mini. These models are made to pause and think through questions before they answer. This makes them great for tasks that need deep thinking and analysis. You can find them in both ChatGPT and the API. This means developers and users get better performance and new features.

Benefits

OpenAI o3 is the most advanced reasoning model so far. It is really good at tasks that need deep thinking in many different areas. It can handle complex tasks like coding, math, and science. One cool thing it can do is think with images. This means it can process and understand visual inputs right away. This is great for tasks that involve real-world objects, diagrams, data visualizations, and complex scenes.

OpenAI o4-mini is a smaller, more affordable model. It offers a good mix of intelligence, speed, and cost-efficiency. It is made to work well and be cost-effective. This makes it good for tasks that need a lot of processing. The model does well in math, coding, and visual tasks.

Use Cases

Both models can be used in many different ways. They are great for complex data analysis, advanced science research, coding and software engineering, education and tutoring, creating and understanding content with different types of media, business intelligence and strategy, and creative problem-solving.

Pricing

OpenAI has set different prices for the new models. The prices show what each model can do and who it is for. The prices are based on 1 million tokens. Tokens are pieces of words.

OpenAI o3 Pricing:
\n- Input: $10.00 per 1 million tokens
\n- Cached Input: $2.50 per 1 million tokens
\n- Output: $40.00 per 1 million tokens

OpenAI o4-mini Pricing:
\n- Input: $1.10 per 1 million tokens
\n- Cached Input: $0.275 per 1 million tokens
\n- Output: $4.40 per 1 million tokens

Vibes

Both o3 and o4-mini models have shown great skills in many standard tests. o3 got a score of 87.7% on the GPQA Diamond benchmark. This test has expert-level science questions that are not found online. On SWE-bench Verified, o3 scored 71.7%, compared to 48.9% for o1. On Codeforces, o3 reached an Elo score of 2727, whereas o1 scored 1891.

Additional Information

Both models can be found on OpenAI\'\'s ChatGPT platform and API services. ChatGPT Plus, Pro, and Team users can use o3 and o4-mini right away. Enterprise access will be available in February. Free plan users can try o3 by choosing \'\'Reason\'\' in the message composer or by regenerating a response. Developers can add o3 and o4-mini to their apps using OpenAI\'\'s Chat Completions API and Responses API. This lets them make custom AI solutions for different platforms.