ChatGPT (flat subscription tiers)
- Paying for access: Plus is $20/month, Pro comes in tiers at $100, $200 and $500/month, and Business Standard seats are $25/user/month ($20 billed annually).
- Same amount is paid whether answers are right or wrong, so quality risk sits with the subscriber. OpenAI bets that the average subscriber uses less than the the price covers, with occasional heavy users breaking that limit, so cost risk falls slightly on the company.
GitHub Copilot (per seat)
- Paying for a person, but now with a meter attached: Business is $19/user/month and Enterprise is $39/user/month. Since 6/1/26, usage is calculated from token consumption, including input, output, and cached tokens, at the listed API rates for each model. Once the credits run out, extra usage is $0.01 per credit if the admin allows it.
- The seat now covers access for one person plus a prepaid usage budget, which moves cost risk and quality risk to the consumer. Bad generated code still costs review time, flat fee, plus additional token fees.
Fin (per resolved customer conversation)
- Paying for an outcome: A resolution costs $0.99, and a sales qualification costs $9.99. User isn’t charged if Fin can’t answer or the customer asks for a human.
- This is the only one of the three where the vendor carries quality risk. When Fin fails, the company earns nothing. If a customer later comes back to a resolved conversation for more help, even in a later billing period, that resolution is deducted.
The model that would best fit my team’s product, Accord, would be GitHub Copilot without the extra token consumption enabled. Since the product is going to be seat-based and usage will be fairly limited and predictable, it makes sense to pay for a base rate for each user accounting for their maximum usage rate.
