Pricing is product. A SaaS product with a bad pricing model will lose to a mediocre product with a good one, and in 2026 the ground under every pricing model is shifting because AI features broke the old unit economics. Inference costs make pure per-seat dangerous on AI-heavy products, and pure usage-based pricing still frightens procurement teams who need predictable budgets. The answer for most teams is a hybrid — a seat or tier base plus metered usage on top — but the right hybrid depends on who uses the product, how they buy, and what actually costs money to deliver. This post breaks down the four pricing models, when each one fits, how Vercel, Linear, and Notion got to their current shapes, and how to migrate between models without detonating revenue.
The four models
| Model | What customers pay for | Best for | Margin risk |
|---|---|---|---|
| Per-seat | Active users × monthly price | Collaboration tools; predictable budgets | Low — revenue scales with users, cost barely moves |
| Usage-based | Metered consumption (API calls, compute, tokens) | Infrastructure, AI, data products | Medium — requires tight cost tracking |
| Hybrid tiered | Base fee + included quota + usage overage | Most SaaS with variable-cost workloads | Low — base covers fixed cost, meter covers variable |
| Freemium + paid tiers | Free up to a limit; paid when limits bite | PLG, viral loops, developer tools | Medium — free tier can be a CAC line item |
Two market facts matter for 2026. Per-seat is still the most common model — around two-thirds of SaaS companies use some seat-based component — but pure per-seat is in decline. Hybrid is the fastest-growing shape; industry surveys put adoption around 43% in 2026 and projecting higher through the year. And the share of SaaS companies with any usage-based component is up meaningfully from a couple of years ago, driven almost entirely by AI features where per-seat pricing would torch the margin.
Per-seat pricing — when it still wins
Per-seat works when the value of the product is the collaboration that happens on it. More users in a Slack workspace means more channels, more conversations, more reason for the next user to join. More editors in Figma means more designs, more reviewers, more files that new joiners need access to. In that world, the seat count is a proxy for the value delivered, and customers accept paying per-seat because the math tracks their own mental model.
Per-seat breaks down in two places. First, when seats are a poor proxy for value — an automation tool where one power user drives 100× the workload of a normal seat, or an analytics tool where most seats just view dashboards. Second, when the underlying cost is variable and tied to usage, not users — an AI feature that costs pennies per inference is fine until a tenant runs a million inferences against their flat seat price.
Linear is the cleaner 2026 example of per-seat done well. Paid plans are per active member, with a generous free tier for small teams. The product is a collaboration tool where value genuinely scales with team size, and Linear's margins are fixed-cost-dominant rather than variable-cost-dominant, so seats track the economics. Notion charges a flat per-active-user monthly price on team plans for similar reasons.
Usage-based pricing — when variable cost rules
Usage-based makes sense when the cost to deliver the product scales with consumption, not with seat count. Cloud infrastructure, AI inference, data processing, outbound email, SMS — anywhere a customer's action maps to a real dollar of COGS, usage-based pricing aligns the revenue line with the cost line and protects margin automatically.
The counterweight: enterprise procurement teams hate it. A variable monthly bill is hard to budget, hard to forecast, hard to approve in a purchase order. That friction kills deals in top-down enterprise sales, and it's why most 'pure' usage-based companies offer committed-spend annual contracts with volume discounts — customers commit to 100k of credit up front at a 15% discount, which converts the experience back to a predictable annual number while preserving the consumption economics underneath.
Stripe's 2026 push into metered billing is the most important infrastructure update for usage-based pricers. The Billing Meter API lets a developer stream granular events — tokens processed, model API calls, agent tasks — into Stripe, which then aggregates and invoices automatically. The advanced usage-based SKU adds multi-dimensional meters, per-attribute pricing, and real-time credit burndown. If you're shipping an AI-heavy SaaS, Stripe Meter plus a thin internal events pipeline removes most of the reasons teams used to build metering in-house.
Hybrid tiered pricing — the 2026 default
Hybrid pricing is the model most SaaS businesses should be on in 2026. It looks like: a tier (Starter, Pro, Business) with a monthly base price, an included quota of some metered resource (API calls, AI credits, events, seats), and overage pricing when the quota runs out. The base covers fixed cost, the meter covers variable cost, the tier lets sales negotiate without breaking the pricing page, and the included quota gives procurement a predictable invoice most months.
Vercel is the prototypical example in 2026. The Pro plan is 20 dollars per user per month with 20 dollars of usage credit included; consumption beyond that (bandwidth, compute, image optimization) meters on top. Customers on predictable workloads see a bill that looks like a per-seat subscription. Customers on bursty or high-scale workloads pay more, and the per-seat piece covers the platform's fixed costs regardless. The shape handles product-led growth teams and enterprise buyers in the same pricing page.
Pick the meter first, the tier second. The meter should be a number the customer can already see, understand, and correlate to value — events, API calls, seats, minutes of video processed. If the customer can't predict their bill before reading it, you priced it in the wrong unit.
Freemium and product-led growth
Freemium works when the free tier is a functioning acquisition channel. The product has to be useful enough at zero dollars that people start using it and tell others, and the upgrade trigger — usually a hard quota, a collaboration limit, or a feature gate — has to hit at the exact point where the user has already felt value and can't turn back. If the free tier is too thin, the viral loop dies; if it's too generous, nobody converts. Notion, Linear, and Figma all spend serious product effort on exactly this calibration.
Treat the cost of the free tier as a line item in CAC. Free users are not free — they consume storage, bandwidth, support cycles. Most freemium products converge on a ratio of paid-to-free that governs the math; a 2% conversion rate with a 50-dollar ACV and a 3-dollar-per-free-user infrastructure cost is a viable unit economic, a 1% conversion rate at the same numbers isn't.
The AI cost pass-through problem
Every AI-heavy SaaS in 2026 is fighting the same margin equation. Traditional SaaS ships with 75–85% gross margin because the marginal cost of an additional user is nearly zero. AI-first SaaS ships with 55–70% — sometimes lower — because every inference has a real COGS in tokens and GPU time. A pure per-seat AI product is a margin time bomb: one power user can burn their entire monthly seat price on a single afternoon of heavy use.
Directly passing AI costs through to customers as a line-item invoice often triggers pushback — customers want a price, not an itemized receipt. The better pattern: bake an included AI quota into each tier, meter the overage, and surface a 'usage so far this period' number in-product so customers know before the bill arrives. Transparency without itemization.
Three concrete moves that work in 2026. First, tier the AI quota — Starter gets enough to try the feature, Pro gets enough for normal daily use, Business gets enough for heavy use, and anything beyond meters. Second, decouple model choice from price where it makes sense — an 'economy mode' using a smaller, cheaper model for bulk tasks lets customers stretch their quota on low-stakes work. Third, bundle value narratively — 'included AI assistant' sells better than 'X million tokens per month'; the tokens are the mechanism, the assistant is the product.
Migrating between models without burning customers
The hardest part of pricing isn't picking a model — it's changing one. Customers who bought on the old pricing feel every change, especially the ones that look like increases. A sound migration follows four steps.
- Grandfather existing contracts. Everyone on the old plan stays on it until their next renewal. No retroactive changes; no surprises on the next invoice.
- Publish the new model alongside the old for a transition window. Lets prospects compare, lets the sales team sell the new one, lets you collect signal on which tiers actually pick up.
- Offer a migration path with a visible win. Move-to-new-plan usually comes with a credit, a quota bump, or a grandfathered discount for the first 6–12 months. The customer moves because it's better for them, not because you forced it.
- Communicate early and in writing. Email, in-app banner, account-manager outreach for mid-market and up. Every customer should know the change is coming weeks before it hits their invoice.
One underrated tactic: annualize the transition. If the new pricing is materially higher for some cohort of customers, offer a 12-month locked rate at the old price in exchange for a prepaid annual. Customers get predictability, the company gets cash, and the actual pricing change rolls out as renewals hit the new model naturally.
Key takeaways
- Per-seat still wins for true collaboration tools where value scales with team size. Linear and Notion earned theirs.
- Usage-based protects margin on variable-cost workloads but fights procurement. Committed-spend annual contracts bridge the gap.
- Hybrid tiered — base plus included quota plus overage — is the 2026 default for most SaaS, and Vercel's shape is the cleanest example.
- Freemium is an acquisition channel with a real CAC cost. Calibrate the upgrade trigger to where value is felt, not where storage is filled.
- AI features break per-seat economics. Tier AI quotas, meter overage, show usage in-product, resist the urge to itemize tokens on the invoice.
- Migrate carefully: grandfather, publish alongside, offer a visible win, communicate early. Pricing changes cost goodwill if you rush.