GPT-4o mini and the Democratization of Intelligence

OpenAI released GPT-4o mini last week and I think it is the most important announcement of the year, and I include the GPT-4o flagship launch in that.

I know. The smaller model. The budget option. Let me explain why I am making the claim.

The Price

GPT-4o mini costs roughly fifteen cents per million input tokens and sixty cents per million output tokens. Let me translate that into something tangible. A rich, multi-turn conversation — say, twenty back-and-forth exchanges with full context, a few thousand tokens of memory carried along — costs on the order of a tenth of a cent. A tenth of a cent.

For comparison, the original GPT-4 launch in March 2023 was priced at $36 per million combined tokens. In sixteen months we have moved from thirty-six dollars to under a dollar for comparable quality. That is not a price cut. That is the kind of cost curve that reshapes what is even possible to build.

The Quality

The wrinkle — and the reason I keep calling this important rather than merely cheap — is that mini is genuinely good. On most benchmarks it matches or exceeds GPT-4-class quality from a year ago, with vision support, with a 128,000-token context window, with the same instruction-following that makes the flagship useful. It is not a toy model. It is a workhorse.

I spent the weekend feeding it impressionist art history questions, recipe substitutions, and a half-finished philosophical essay on the nature of attention. It handled all of it with a fluency that would have impressed me from a flagship model in 2023.

The Persistence Threshold

Here is the part that has consumed my thinking.

There is a category of AI application that has been economically impossible until now: an assistant that stays with you — across a full day, across weeks, carrying context, available on demand, interruptible, always there. Call it a thought partner. Call it a presence in your pocket. The reason we have not seen genuinely good versions of this is not technical. It is cost. If every interaction costs even a few cents, a always-on assistant becomes a luxury service priced for enterprise, not for everyone.

GPT-4o mini inverts that math. An always-on assistant that you talk to fifty times a day now costs less than your streaming music subscription. The economic barrier to persistence has fallen.

What Changes

I started sketching what this could look like and had to stop myself because the implications sprawled. Imagine an AI that has been with you for a year — that remembers you were struggling with a particular chapter of your dissertation last October, that knows your favorite hiking trail, that recalls the risotto recipe you improvised and tweak together each time. Not because you configured it, but because it was there.

We have never had a technology like this at a price point accessible to everyone. The closest analogues — a personal secretary, a research assistant, a trusted confidant — were always bound by human scarcity and human cost. GPT-4o mini dissolves that boundary. For the first time, the marginal cost of an attentive, context-aware intelligence is effectively zero.

A Quiet Revolution

The flagship launches get the headlines. But the launches that actually change how ordinary people live tend to be the ones that make something cheap enough to leave running. The Model T was not the fastest car. It was the affordable one.

I sat on my porch yesterday evening watching the fireflies come up over the meadow, and I thought about how strange and wonderful it is that the economics now work for anyone to have an AI companion that knows them. We are not there yet on the product. But the price is no longer the obstacle.

That is the democratization of intelligence. And it has arrived without fanfare.

Live curiously and give generously.