Is usage-based pricing becoming the norm for AI tools?

We're in the final stages of private beta testing for our new product and making it "AI-native" has yielded a lot of the same thinking - the TCO of the entire stack is something you have to watch closely because the product leaders out there are using VC-subsidized tokens and infra costs to offer unrealistically expensive AI workflows (think document processing/storage, image generation, memory management, context compression/summarization, etc) and those margins are not sustainable for sure.

To answer your questions:

• We are sticking with SaaS pricing - or at least, we intend to! - but with the option to "Allow me to continue working" add-ons to top up accounts that need more usage.

• I don't think we can avoid Hybrid model territory once you have individual tiers vs organization/team tiers (like we are planning) because usage patterns scale differently when there are work teams/depts involved.

• Don't know yet - we launch next month!

My prediction is that if you can balance quality of outcomes in your AI-assisted domain with a reasonably priced plan that a typical user can live on for a month without exhausting limits in a few days, then you've found an approach that will generate sticky users.

I think hybrid pricing is probably where a lot of AI tools are heading. Pure monthly SaaS feels clean, but AI changes the math because every prompt, generation, video, image, or deep research task can have a real cost behind it.

I’m building Traction, and we’ve been thinking about this too. Some parts of the product should feel like normal subscription value: planning, leads, follow-up, visibility, analytics, business context, etc. But heavier AI media generation probably needs credits or usage limits because the cost structure is different.

The hard part is making it feel fair instead of annoying. Users hate feeling nickel-and-dimed, but founders also cannot pretend compute is free. My instinct is: base plan for the core platform, included monthly credits for heavier AI features, then optional credit packs for people who need more volume.

Running into this exact tension with Sharpread. Every analysis costs real money in Anthropic API calls plus EDGAR data fetches, so pure flat-rate SaaS feels like a race to the bottom on margin.

We landed on a hybrid: a fixed number of analyses per tier rather than pure usage billing. Users get 20 analyses for $5, which is simple to understand and predictable for them, while we can model our costs reliably on the back end.

The thing I have noticed: usage-based pricing creates anxiety for users who are not sure what they will be charged at the end of the month. A fixed credit model gives them the same flexibility without the open-ended cost fear. Conversions felt cleaner once we moved away from anything that looked like a meter running.

The real architectural pain is metering and enforcement, not the pricing model itself. That work is the same whether you charge per use or bundle credits into tiers.

wow, great conversation happening here ---- from my perspective (working in the VC industry with a lot of founders):

Im starting to see some great SaaS companies moving to "consumption" pricing with a small SaaS fee - depending also in what kind of problem are they solving and how painful/hard is to switch, there are a lot of sectors that are being "commoditized" today

If I wanted to start changing the model, I would begin talking to my clients and understanding their needs at least three months before renewal

usage-based is not really a pricing question, it is a forecasting question your buyer has to answer. the moment a finance team cannot predict next month's bill within a tight band, the tool moves from line-item to procurement review, and the sales cycle doubles. the hybrid that works is a flat platform fee that covers predictable workload plus metered overage on the genuinely variable bit, usually long-context or agentic loops. one thing worth instrumenting before you change pricing: cost per successful outcome, not cost per token. token bills go up when your prompts get worse, and you do not want to charge customers for your own regressions.

Personally, I don’t really like the pay-by-volume model, because I’m afraid of worrying about the volume. I’ve been using AI for a long time, and as long as there’s a monthly subscription that clears the quota regularly, I will choose it without hesitation. Believe me, as long as there are enough users, the consumption will be evenly shared, but the prerequisite is real users, monitor it well, and don't abuse it.

I think the shift is real but not universal. The systems that genuinely cannot sustain flat subscriptions are the ones where LLM inference is the primary cost driver. One heavy user costs you 10x more than a light one, and flat pricing just transfers that risk onto you. For those, usage-based or hybrid is not a choice, it is a math problem.

For everything else it depends on whether your cost per query is variable enough to matter at scale.

I am building KBrain.io, an MCP-based knowledge marketplace. My intuition is that subscription works for us because the core cost is retrieval, not inference. But I have not fully modeled what one query costs at 10k/day versus 100k/day, and I think that is the honest answer most people in this thread are sitting on too. Pricing strategy is downstream of cost structure. Until you instrument and model it, any model you pick is a guess dressed up as a decision.

I started with the subscription model and quickly added bundled roll-over credits. So many AI tool users want something free, but the next best thing is a payment model that they can minimize, or feel in control. This is a feature, that many users are grabbing on to. Noone wants another subscription, so we provide the AI Creative services of several platforms in one spot, and allow for pay as you go credits. So far the buy in has been 50/50 sub vs bundled, but having both options is effective. See our credit model: https://aetherwavestudio.com/buy-credits

Hybrid pricing feels inevitable for AI tools: a base plan for predictable value, then usage for expensive workloads. The key is making the meter understandable. If users can’t predict what costs money, they’ll either churn or underuse the product.

Sadly you may be right... this means still more work for our product :( never ending chase!