The article argues that most AI chat app pricing models are unsustainable because they hide complex costs like retries, context growth, and multi-model calls behind flat subscriptions, and suggests usage-based pricing with limits and overages is more viable.
Over the past year, partly because we work on MoR/payment infrastructure for ai saas companies, I’ve ended up talking to a lot of teams building AI chat products. And one thing keeps standing out to me, most of the pricing makes absolutely no sense once you look under the hood. Almost everyone starts in the same place. A simple monthly subscription, “unlimited” usage somewhere on the landing page, maybe a higher tier for power users. It looks clean and competitive, and honestly I understand why teams do it. But then the product gets more sophisticated. One user message stops being one model call. There’s retrieval happening, memory systems, retries, summarization, tool calls, sometimes multiple models involved in the same workflow. From the user’s perspective it still feels like “I sent one message.” Internally it can turn into half a dozen billable operations. That gap is where I keep seeing teams get hurt. The other thing that catches people off guard is context growth. A customer keeps using the same chat thread for months, the product keeps feeding more history back into the model, and suddenly the cost per interaction quietly multiplies without the experience changing much for the user. Retries are another hidden one. Providers get flaky, requests retry automatically in the background, and costs spike without anybody immediately realizing why. A lot of teams don’t even have good visibility into how much of their bill is retry traffic versus real usage. The whole thing reminds me a bit of early ISP pricing. Flat subscription on the surface, wildly variable infrastructure cost underneath. And lowkey, after seeing enough of these companies up close, I’ve started thinking “unlimited AI chat” is mostly a temporary phase. The economics just get weird once heavy users show up. The teams that seem healthiest financially usually land in the same place eventually: some kind of fixed subscription with usage limits or overages layered underneath. Not because it’s exciting pricing, but because it’s the only thing that consistently survives contact with real usage patterns. We got this wrong too at one point. Had a pricing tier that looked completely reasonable until a small group of users started running agent-heavy workflows through it and quietly destroyed the margins for months before anyone fully noticed. The fix ended up being the boring stuff, quotas, usage alerts, overages. Not very exciting from a product perspective, but a lot more sustainable. Curious how other teams are thinking about this now, especially as products get more agentic. Are people still trying to hide all the underlying complexity behind flat pricing, or are users getting more comfortable with usage-based models now?
The article critiques the restrictive nature of current AI pricing models, highlighting how daily quotas and stacked limits hinder productivity and user trust.
The author argues that current AI usage pricing models are broken because users are charged for hidden 'thinking' tokens that are not visible to them, creating a trust-me billing system. They propose that labs should either adjust output token pricing or bill explicitly for compute.
An analysis of how AI agent subscription services manage API costs, revealing that pooling, usage caps, and first-party model ownership are key factors, while many resellers operate on thin or negative margins.
Analysis of Claude Fable 5's cost and pricing model, Anthropic's decision to stop including frontier models in subscriptions and move to per-token pricing, and the broader economic implications for AI access and inequality.
The article argues that current AI subscription pricing is heavily subsidized by providers like OpenAI, Anthropic, and Google, creating a ticking time bomb for enterprises that build workflows on artificially low costs; when prices correct, organizations face massive cost increases.