What OpenAI's new reasoning model o3 means for indie hackers

• OpenAI announced o3, a new reasoning model, which approaches AGI-levels of performance on benchmarks.

• The model won't be available until January 2025 or later.

• Founders should use reasoning models to make long-term predictions, while building nimble businesses that can earn short-term profits.

AI just leveled up again.

OpenAI just announced o3, their next-generation reasoning model.

This is the last and biggest of its "12 days of shipmas" announcements, and I've been thinking about what it means for us as founders.

What is o3, and why is it a big deal?

o3 is a new family of reasoning models, which will include o3 and o3-mini. And, if they follow in o1's footsteps, possibly o3-pro-mode at some point.

For those just catching up, a reasoning model is a popular new type of AI model that's capable of, well, reasoning. They can walk step-by-step through the problems you give them and the questions you ask, devising a plan and checking their own work as they go. As a result, they take longer to respond (sometimes much longer), but they can more accurately solve more complex problems. They even tend to write better.

OpenAI's first publicly-available reasoning model, o1, was released in September of this year. Since then, OpenAI, Google, and others have released improved reasoning models.

I've personally been paying OpenAI $200/month (yikes) for access to their most advanced o1 model, o1-pro-mode, just to test it out and see how well it writes and reasons.

It's spectacular. I think 6 months from now we'll wonder how we ever settled for less than the reasoning models.

So just how much better is o3 than the current generation of o1 models?

Let me simplify this for you:

• ARC-AGI is a benchmark meant to measure our progress toward AGI, i.e. how good an AI is at acquiring new skills outside its training data.

• To match human performance, an AI would need to score about an 85%.

• o1, OpenAI's previous reasoning model, scored between 25-32%.

• o3 has made a massive leap forward, reaching scores of 87.5% at its best, and generally performing at least 3x better than o1.

Is it AGI? Not quite. But in many cases it's virtually indistinguishable, and that should cause every founder to stop and think about what this means for us.

What founders should do about AGI

Just a couple days ago, writing coach David Perell tweeted about just how far AI has come:

I responded, "AI is coming for speed, quality, creativity, clarity, consistency, efficiency, style, adaptability, coherence, depth, brevity, variety, novelty, elegance, wit, playfulness, analysis, synthesis, and originality. The only value left will be authenticity."

And if AI can become a better writer than writers, the smart bet is that it can become a better founder than founders in time.

o3 isn't out quite yet — they're doing safety testing on it. And if I had to guess, I'd estimate it won't become available to the public until late January or early February. So now is our time to prepare. Not just for the o3 launch, but for the rise of more advanced reasoning models and AGI-like performance in 2025 and beyond.

My biggest takeaways for indie hackers:

You probably can't predict the future, but you should try. As Sahil Lavingia recently tweeted, "Build for 2030." You should absolutely take a few minutes to try and look ahead. Appropriately, I recommend using o1 to help. If these new reasoning models are good at anything, it's planning.

Here's a prompt I used that produced some interesting results, and is helping me get into the mindset of building for the future:

I run Indie Hackers, a media company and community for people founders who are trying to run and grow online businesses. We currently charge a subscription fee for access to unlimited premium content (niche news and how-to) written by our journalists. I'm curious about how ongoing AI development might affect my business going forward, as AI reasoning models get more powerful, agents get more powerful, LLMs improve at writing and it becomes indistinguishable from humans, we get closer to AGI, and other developers and companies build on top of this.

Please examine 5 of the most likely second-order effects that will occur in the short-term that are relevant to my business. Then, for each of those, extrapolate out a bit and predict a third-order effect that's relevant to my business, maybe 1-2 years away. Finally, for each of those, extrapolate again and predict a fourth-order effect that's relevant to my business, maybe 3-5 years away.

When you extrapolate each time, state how the world/markets/environment/technology/demand/customers will change, and elaborate. Only then discuss how my business might adapt or change, thrive or suffer.”

Like with all prompts, you need to tweak it for your own business, and tweak it again if the AI doesn't give you the response you're looking for. But a few back-and-forths tweaking prompts is more than worth the insight you'll get.

Plan together with o1 and (eventually) o3. If you haven't caught on yet, if there's one thing reasoning models are good at for founders, it's planning. If you make a plan for a new idea, a new feature, a marketing strategy, or anything else, copy-paste it into o1 and ask what it thinks, or whether it can spot any flaws or . The best founders I know are habitually using the latest AI models as executive coaches, and workshopping how to get as much of their business into the limited context windows as possible.

This post from Sam Parr has a lot of value not just in the responses, but in his description of his process:

Finally…

Move fast and build for now. I know I just said to look ahead to 2030, but in a world where change is frequent yet uncertain, being nimble and reacting to changes faster than others is a super power. And if indie hackers can do anything better the rest of the startup community, it's move fast.

For example, as soon as image models got good, Pieter Levels and Danny Postma capitalized by building PhotoAI and Headshot Pro, both of which have earned them millions of dollars. AI resume builders, business plan builders, and other practical tools have done well, too.

My advice here is to stay away from general purpose tools, and instead consider using o3 and other reasoning models to build better AI versions of practical businesses that have generated revenue for decades but required humans in the loop.

Since o3 is all about taking more time to reason more deeply, consider businesses that require thought and research, like travel planning, grant writing, and market research.

Will these businesses be around in 2 years, 5 years, 10 years? Maybe not.

But they can do well in the short term, because excitement about new AI capabilities fuels customer curiosity, Google searches, and press coverage.

Market what you build as, "Move over XYZ! New o3-powered tool does XYZ better than humans ever could," pitch the press to give them a good story, and focus on SEO from there if your niche caters to it.

And if you strike it big, come back to Indie Hackers and email us so we can share your story.

Happy building!