How to price AI features in 2026: credits, seats or outcomes?

Seat-based pricing assumed software costs almost nothing to run per user. AI features break that assumption: every request has a real cost, and your heaviest user can cost a hundred times more than your lightest.

So how are SaaS companies pricing AI in 2026?

What the market is doing

• Around 60% of SaaS companies now use some form of usage-based pricing, and about 46% run hybrid models, according to OpenView's pricing research
• A 2026 OpenMeter survey of 230 companies found hybrid pricing at 37%, AI credits live at 29% with another 33% planning them, and outcome-based pricing at 5% today, with 31% expecting to use it within three years
• The same survey put typical gross margin on AI usage at around 50%, well below classic SaaS
• Big platforms have moved too: GitHub Copilot shifted to AI credits in June 2026, and Intercom's Fin has long charged $0.99 per resolution

The four models

Model	How it works	Best for	Risk
Bundled in seats	AI included in existing plans	Light, low-cost features such as summaries	Heavy users eat your margin
Hybrid allowance	Plan includes X AI units, then overage or top-up	Most B2B SaaS	Allowance design takes iteration
Credits	Prepaid or included credits spent per action	Varied AI actions with different costs	Customers find credits abstract
Outcome-based	Pay per resolved ticket, booked meeting and so on	Agents with measurable results	Attribution disputes, revenue volatility

Start with unit economics

Before choosing a model, know your numbers per AI action:

1. Cost per action: tokens in and out × model price, plus retrieval, tools and infrastructure
2. Distribution: median versus 95th-percentile usage per account
3. Value per action: time saved, revenue created or cost avoided for the customer

Then reduce cost before you set price. Model routing, caching and prompt trimming often cut cost by 50–80%; see our LLM cost optimisation guide and model routing guide.

A decision framework

Choose bundled if the 95th-percentile user costs less than about 5% of their subscription.

Choose hybrid if usage varies a lot but customers want predictable bills. Set the allowance so 70–80% of accounts never hit it, and make top-ups painless.

Choose credits if you have many AI actions with very different costs, such as text, image and agent runs. Publish a clear credit table and show balances in the product.

Choose outcome-based if your AI completes a job end to end, and you can measure success unambiguously. See our guide to measuring AI support agent ROI for what "resolved" should mean.

Implementation checklist

• Metering: log every AI action with account, feature, model, tokens and cost
• Real-time balances: customers see usage before they're surprised by it
• Soft limits and alerts at 80% and 100% of an allowance
• Graceful degradation: fall back to a smaller model rather than hard-blocking, where it makes sense
• Billing integration: Stripe usage-based billing, or metering tools such as OpenMeter, Orb or Metronome
• Admin controls: let B2B customers cap spend per workspace or team

Communicating it

• Price in customer language ("100 AI drafts per month"), not tokens
• Show an example bill for a typical customer on the pricing page
• Grandfather existing customers for a period when you introduce AI charges

Bottom line

For most startups in 2026, the answer is hybrid: a predictable plan with a generous AI allowance, metered properly underneath, with credits or outcomes once you understand your usage curve.

Need metering, credits or billing built? Our custom software team ships pricing infrastructure in sprints. Let's talk.