The 60-second answer
In 2026 the SaaS pricing landscape has converged on five credible patterns. None of them is universally best. Pick the one whose billable axis matches the customer’s perceived value axis:
- Tiered — good for early-stage products with limited usage data. Three tiers, anchor pricing in the middle.
- Per-seat — default for collaboration tools (Slack, Notion, Linear). Scales with team size.
- Usage-based — default for infrastructure (Twilio, Vercel, Snowflake). Scales with consumption.
- Hybrid (seat + usage) — the 2026 default for AI products. A base seat fee plus metered AI credits.
- PLG/freemium — only for products with viral loops or content leverage.
Pick the right value axis first
Pricing is a function of one decision: what unit does the customer mentally use to measure value? For Slack, it is “people in the workspace.” For Twilio, it is “messages sent.” For Snowflake, it is “queries executed.” For Cursor in 2026, it is “AI completions consumed.” If your billable axis matches the customer’s mental value axis, pricing is invisible. If it doesn’t, every renewal becomes an argument.
| Product type | Customer value axis | Recommended billing axis |
|---|---|---|
| Team collaboration tool | Number of collaborators | Per-seat |
| CRM / sales tool | Pipeline value × contacts | Tiered + contact volume |
| Email / messaging API | Messages delivered | Usage-based |
| Vertical SaaS (e.g., HVAC ops) | Revenue managed × users | Tiered with seat floor |
| AI copilot / coding tool | Hours saved / completions accepted | Hybrid (seat + AI credits) |
| AI agent / autonomous workflow | Tasks completed end-to-end | Outcome-based or per-run |
Tiered pricing — the workhorse
Three tiers (Starter / Pro / Business) with feature gates between them. The middle tier is the anchor; design the price page so 60–70% of customers self-select into it. Tiered pricing is the right starting point when you don’t yet have usage data to design metered billing.
Typical 2026 mid-market B2B SaaS tiered prices:
- Starter: $29–59/month. Feature-gated, no admin controls, community support.
- Pro: $79–199/month. The intended default. Real admin features, integrations, email support.
- Business: $299–799/month. SSO, audit logs, SLA, priority support.
- Enterprise: custom. SAML, DPA, EU data residency, dedicated CSM.
Common mistake: pricing the Pro tier too close to the Starter. The right spread is roughly 3× between Starter and Pro and 3–4× between Pro and Business. Anything narrower and customers default to the cheapest option.
Per-seat pricing — the collaboration default
Per-seat works when adding people to your tool genuinely creates value — collaboration tools (Slack, Notion, Linear, Figma, GitHub), CRMs (HubSpot, Salesforce), and project management (Asana, Monday).
The seat price in 2026 mid-market collaboration ranges from $8–15/seat at the basic tier to $18–30/seat at the standard tier and $45–90/seat at the business tier. Below $8 you cannot sustain CAC; above $90 procurement triggers an RFP.
The big risk of per-seat: customer optimisation. Buyers buy fewer seats than they have users by routing access through shared accounts. Counter this by enforcing SSO + audit at the business tier (the same tier where per-seat starts to bite).
Usage-based pricing — infrastructure and AI
Usage-based pricing aligns customer cost to customer value when the unit of value is consumable. Twilio per message. Stripe per transaction. AWS per GB. OpenAI per token. The 2026 reference prices for the major frontier LLMs are illustrative:
| Model (2026) | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| Claude 4.6 Opus | $15 | $75 |
| Claude 4.6 Sonnet | $3 | $15 |
| GPT-4o | $2.50 | $10 |
| o3 | $10 | $40 |
| Gemini 2.5 Pro | $1.25 | $5 |
| Mistral Large 3 | $2 | $6 |
| DeepSeek V3 | $0.27 | $1.10 |
| Llama 4 (Bedrock) | $0.90 | $2.70 |
The 5–50× spread across model tiers means “cost per AI feature” depends entirely on routing strategy. We covered this in detail under generative AI integration and AI agents development.
Usage-based pricing requires real metering infrastructure: an event pipeline, idempotent counters, customer-facing usage dashboards, and a billing engine like Stripe Billing’s metered subscriptions, Orb, Metronome or Lago. Budget 4–8 weeks of engineering for production-grade metering.
Hybrid pricing — the 2026 default for AI SaaS
The breakout pattern of 2025–2026. A base seat fee covers fixed costs, support and the product baseline. A separate metered component covers variable AI inference cost. Customers get predictable budgeting on the seat side and pay-for-value on the AI side.
Reference 2026 implementations:
- Cursor — $20/seat/month with included “fast” credits and metered overage for premium model use.
- Linear with AI — $14/seat plus AI credits for agent runs.
- Notion AI — $10/seat add-on with effectively-unlimited usage but throttled to non-frontier models.
- HubSpot Breeze — tiered seat license + AI credits per workflow run.
The hybrid model gives you a defensible gross margin on the seat (75–85%) and a passthrough-plus-margin (20–40%) on the AI credits. Most importantly: a customer who suddenly 10×’s their AI usage does not blow up your COGS — they pay for it.
PLG and freemium — viral loop or trap?
Product-led growth works when one of two conditions holds: (1) free users invite paid users (Slack, Loom, Calendly, Figma), or (2) free users generate signal that improves the product (analytics, search, marketplace). Without one of these, freemium is a giant cost centre.
A few honest numbers from 2026 PLG benchmarks:
- Free-to-paid conversion in healthy PLG SaaS: 2–5% within 90 days.
- Below 1%, your free tier is a charity.
- Above 8%, your free tier is too constrained — you are leaving viral pipeline on the table.
- Support cost per free user in collaboration SaaS: $0.40–1.20/month. Multiply by your free user count to see the bill.
Enterprise pricing — talk to sales, but smartly
“Contact sales” for enterprise is still the right pattern in 2026, but lazy execution loses deals. Best-in-class enterprise pages now publish:
- The capabilities (SSO/SAML, DPA, audit logs, EU data residency, custom DPA, BAA on request).
- An indicative starting price (“starts at $30k/year”).
- A buyer-friendly procurement kit (security pack, sub-processor list, ISO/SOC reports under NDA).
Hiding the starting price is a 2018 tactic that now costs deals; mid-market procurement teams filter out vendors that can’t commit to a floor.
AI gross margin and pricing implications
Traditional SaaS hits 75–85% gross margin on hosting + support. AI SaaS in 2026 lands at 55–72% gross margin because inference is genuinely expensive at scale. The implications for pricing:
- Route aggressively. Use Claude 4.6 Sonnet or GPT-4o for the bulk of traffic, escalate to Opus or o3 only for hard cases. Routing saves 40–70% of inference cost.
- Cache aggressively. Prompt caching with Anthropic (90% discount on cached tokens) or OpenAI (50% discount) typically saves 30–55% of input cost on repeat-pattern workloads. See GenAI integration.
- Distil where possible. Fine-tune a small open model (Llama 4 8B, Mistral 7B) on logged Claude/GPT outputs for narrow tasks. 5–15× cheaper inference at 92–97% quality retention.
- Meter the customer. Pure flat-rate AI pricing is suicidal at scale. Even “unlimited” plans must have soft caps and fair-use language.
- EU AI Act note. If your product offers high-risk uses (recruitment, credit, education) you also pay for documentation, logging and human-oversight infra. See EU AI Act compliance for what to budget.
Decision matrix
| Situation | Recommended model |
|---|---|
| Early-stage, no usage data, B2B | Tiered (3 tiers) |
| Collaboration tool, multi-user value | Per-seat with tier modifiers |
| Infrastructure, API-shaped product | Usage-based with monthly minimum |
| AI-heavy product, mid-market | Hybrid (seat + AI credits) |
| Consumer product with viral loop | Freemium → per-user paid tier |
| Vertical SaaS, regulated buyer | Annual contracts, seat floor, tier |
| Autonomous agent / outcome-driven | Outcome-based or per-run |
Six packaging traps
- Anchoring on competitor pricing. Competitors are wrong as often as right. Use Van Westendorp on 30 target customers instead.
- Free tier with no metering. Within 6 months you have power free users costing more than paid customers earn.
- Per-seat on an AI-heavy product. A 5-seat team running 1,000 agent invocations per day per seat destroys your margin overnight.
- Hidden enterprise pricing. Procurement filters you out at the first round.
- Yearly billing discount > 25%. You are training customers that the monthly price is fake. 15–20% is the right anchor.
- Price changes without grandfathering. Existing customers feel betrayed. Always grandfather for at least 12 months, communicate the change 60+ days in advance.
FAQ
What is the best SaaS pricing model in 2026?
None universally. Per-seat for collaboration; usage for infrastructure and AI; hybrid for AI SaaS; tiered as a sensible default for early-stage B2B.
How do I price an AI-heavy SaaS product?
Decouple AI cost from seat price. Base seat fee plus metered AI credits is the 2026 default (Cursor, Linear, Notion all converged here).
What gross margin should a SaaS aim for?
Traditional SaaS 75–85%. AI-native SaaS 55–72%. Below 55% investors price you as a passthrough.
Should I offer a free tier?
Only if there is a viral loop or content/SEO leverage. Otherwise a 14-day trial converts better.
How often should I raise prices?
Once a year on new customers; grandfather existing customers for at least 12 months.
What is the most common SaaS pricing mistake?
Pricing too low. Most SaaS we audit is 30–100% under-priced relative to delivered value.
Re-price and re-engineer with confidence
We help product and finance leaders redesign pricing, build the metering infrastructure, and ship the change — without breaking existing customers.
Last updated 26 May 2026.
