ChatGPT turns language models into a consumer product

One million people signed up in five days for a free chat interface on an existing model. The product layer now decides who captures the value.
ChatGPT crossed one million users on 5 December, five days after OpenAI released it as a free research preview on 30 November, and the model underneath is not new. It is fine-tuned from a model in the GPT-3.5 series, which finished training in early 2022.
The novelty lies in the tuning and the interface, and that shifts the value of language models from whoever trains them to whoever owns the product. Reinforcement learning from human feedback, the same method behind InstructGPT, and a dialogue format turned a tool for developers into something anyone can use without instructions. The model answers follow-up questions, admits mistakes and turns down inappropriate requests. Adoption came within days instead of years.
The distance between a raw model and a product people love turns out to be shorter than the industry assumed: careful tuning, a simple interface and fast iteration. Language models are now usable by people with no technical background, for drafting, coding, analysis and learning. Every conversation is training signal for the next version, so value moves toward whoever owns the user and the feedback loop, as it did when image models went open.
The bill is the other half of the story. Sam Altman put the average cost at single-digit cents per chat and said OpenAI will have to charge for it at some point. Across millions of users that is a large compute bill, so the economics depend on cheaper inference and a paid tier, the model coding assistants proved this summer. The first reliable revenue goes to the cloud that supplies the GPUs, here Microsoft, whose Azure supercomputing infrastructure trained the model. A founder should price every interaction from day one, because a free tier at this cost per use is an expense, not a growth channel.
Accuracy is the weak point. OpenAI itself says the model sometimes writes plausible but wrong answers, so the strongest openings lie in fields where output can be checked and where a company's own data and experts make the model reliable.
Curiosity can fade as fast as it arrived, and retention will decide whether this is a product or a demonstration: whether professionals keep using it for real work week after week. Pricing comes next, both how OpenAI charges for ChatGPT after the preview and whether the same model reaches developers through the API. Rivals will move quickly too, and the speed at which other large labs, Google first among them, put their own conversational models in front of the public will show how long a lead in the product layer lasts.
ChatGPT is the first mass consumer product built on a large language model, and it shows that the product layer can capture value the model alone does not. Fanaura looks for teams that pair a capable model with a product people return to every day and that own the feedback that makes it better, the core of its venture thesis. The model can be rented; the daily habit cannot.
Building an AI-native company? Apply in five minutes. →
More from Fanaura
Building products on language models for professional work where accuracy can be checked? Send us the project.
Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.
Get the news by email
One short note each weekday. No marketing, and you can unsubscribe at any time.