What are the full o1 model and ChatGPT Pro, and who needs them?

Yesterday, December 5, 2024, OpenAI made two announcements in a livestream that opens a series of daily launches it calls "12 Days of OpenAI." First, o1, its reasoning model, left preview: the full version now replaces o1-preview in ChatGPT for paying subscribers. Second, OpenAI introduced ChatGPT Pro, a new plan at $200 a month, ten times the price of ChatGPT Plus.

The two are linked. Pro's main attraction is o1 pro mode, which lets o1 spend even more computing power on each question. But you don't need Pro to use the full o1 itself; every paid plan gets it.

What was released

	Full o1	o1 pro mode	ChatGPT Pro plan
What it is	The finished version of OpenAI's reasoning model	o1 given more computing time per answer	A subscription tier
Who gets it	All paid ChatGPT users	Pro subscribers only	Anyone who pays $200 a month
Main change	Faster, reads images, fewer major errors	More reliable on the hardest problems	Unlimited o1, o1-mini, GPT-4o and Advanced Voice

OpenAI says the full o1 makes 34% fewer major errors than the preview on difficult real-world questions, in its own testing. It is also trained to keep its thinking shorter, which speeds up replies, and it can now reason about pictures you upload, something the preview could not do. TechCrunch notes one surprise: on a few common benchmarks, including MLE-Bench, a test of machine learning engineering tasks, the full o1 scores lower than the preview.

How a reasoning model works

Most chat models start writing their answer immediately, one word after another. A reasoning model like o1 first works through the problem step by step: it breaks the question into steps, tries an approach, checks its work and backs up when something doesn't fit. Only then does it write the reply you see.

Think of two students taking a math test. One writes the first answer that comes to mind. The other uses scrap paper, works it out and checks the result before writing it down. The second is slower, and far more often right on hard problems. On a simple question, like the capital of France, the scrap paper adds nothing.

Compute is the computing power a model uses. o1 pro mode gives o1 more compute for each question, so it can think longer. OpenAI tested this with a strict rule it calls "4/4 reliability": a question counts as solved only if the model gets it right four times out of four. By that standard, OpenAI reports that pro mode beats both o1 and o1-preview on demanding tests in math, science and coding. External expert testers, it says, found its answers more reliably accurate and complete in areas such as data science, programming and legal case analysis. OpenAI also told TechCrunch that pro mode cut errors by 75% on easier coding competition questions.

Because pro mode thinks longer, ChatGPT shows a progress bar while it works and sends a notification if you move to another conversation.

Who ChatGPT Pro is for

Speaking in the livestream, Jason Wei of OpenAI's technical staff said that Pro is meant for power users already pushing the models to their limits in math, programming and writing. OpenAI's own description names researchers and engineers who rely on this kind of help daily.

Here's a quick way to see where each tool fits:

Task	A sensible choice
Drafting an email, summarizing an article	GPT-4o, fast and capable
A tricky spreadsheet formula, a logic puzzle, debugging code	o1, in Plus or Team
A hard proof, a long technical analysis you will check line by line, every day	o1 pro mode, if the time saved is worth $200 a month

OpenAI is also giving 10 Pro subscriptions as grants to medical researchers at US institutions, including scientists at Boston Children's Hospital, Berkeley Lab, Boston University and the Jackson Laboratory, with plans to expand the program.

Why it matters

The full o1 moves reasoning models from experiment to everyday product. Anyone on a paid plan can now pick it from the model menu, attach a photo of a diagram or a handwritten equation, and get a worked answer.

Pro marks a new step in how AI is priced. Until now, consumer chatbot plans have mostly sat around $20 a month. A $200 tier signals that OpenAI sees a group of customers for whom extra thinking time is worth paying for, and that heavy use of reasoning models is costly to run. TechCrunch calls Pro a likely tough sell for all but the most devoted users. OpenAI says it plans to add more compute-intensive features to Pro over time, and it has said o1 is coming to its API, the service developers use to build apps, with function calling and image input.

Try it: If you have ChatGPT Plus, ask the same multi-step question to GPT-4o and to o1. For example: "A train leaves at 9:40 and travels 210 km at 84 km/h, with a 12-minute stop halfway. When does it arrive?" Compare the time each takes, the answer, and how clearly each explains the steps.

Should you use o1 in ChatGPT Plus or pay for Pro?

For most people, Plus is enough. It now includes the full o1, which handles the kind of hard questions that tripped up earlier models: multi-step math, careful logic, debugging and planning. Use o1 when accuracy matters more than speed, and switch back to GPT-4o for quick writing and everyday questions.

Consider Pro only if three things are true: you hit usage limits regularly, your work depends on hard technical problems almost every day, and one avoided error or a few saved hours each month would cover $200. A researcher checking proofs or a developer working through difficult code might fit. A student or casual user almost certainly doesn't.

Whichever you choose, treat a reasoning model's answer as strong work to review, not as a guaranteed result. Ask it to show its steps, check the parts you can, and test its conclusions on a small case you already know the answer to.