ChatGPT charges for access. You pay the same amount whether the answers are good or bad. The tiers are really about how much you can use it, so the price goes up for heavier users. If it gives a wrong answer, you carry that risk entirely. OpenAI carries the cost of running it, up to your plan’s limits, and the higher tiers are how it passes heavy usage back to you.
Copilot charges per person. A company buys a seat for each developer regardless of usage frequency. Even if Copilot suggests bad code, the company still pays. Initially, running costs were handled by GitHub. However, that changed from premium requests to usage-based AI credits. Therefore, heavy use of expensive models costs more.
Fin charges for an outcome. You are not charged when Fin can’t answer and escalates to a human, when a customer asks for a human, or when a procedure fails. In this case, the vendor being Fin carries the highest risk.
For our team project, I would opt for per-seat pricing, but only for people who view the digest (mostly PMs). Everyone who only answers pings or has their work summarized would be free. Usage of AI features would be capped per organization.
