How B2B SaaS companies are pricing AI in 2026

Something significant is happening to B2B SaaS pricing right now, and it is happening faster than most people realise.

AI capabilities are forcing the most fundamental rethink of SaaS pricing models in a decade. The conventional narrative is that the seat-based subscription, the dominant commercial architecture of the last fifteen years, is being replaced by new models better aligned with how AI creates value.

The reality is more nuanced. The seat is surviving as the base commercial relationship, but companies are layering consumption and outcome-based pricing on top of it to capture AI value that a flat per-seat rate cannot reach. Pure seat-based pricing is what is under pressure; the seat itself is not going anywhere. Some companies have developed coherent commercial architectures. Many are still experimenting with how to monetise AI while preserving adoption and simplicity.

This benchmark reviews the publicly available pricing of more than 25 B2B SaaS products that have added meaningful AI capabilities, sourced from public pricing pages in May 2026. The market has split into five distinct commercialisation strategies, and which one a company chooses tells you a lot about how it thinks about what its AI is worth.

This is the most structurally interesting development in the market, and it is concentrated in customer-facing agentic AI. When an AI takes autonomous actions, resolving support tickets, booking meetings, processing requests, charging per seat makes little sense. You are not selling access to a tool. You are selling a result. The leading players in customer support software have moved decisively here.

HubSpot (Breeze Agents) has built the most deliberately layered AI commercial architecture in this benchmark. The base sits on seat subscriptions, with every plan including a monthly allocation of HubSpot Credits: roughly 500 on Starter, 3,000 to 5,000 on Professional, 10,000 on Enterprise. Credits power the supporting AI layer, research prompts, buyer-intent tracking and workflow automation at around 10 credits per action, and are purchasable at $10 per 1,000. On top of that, HubSpot switched its two core agents to outcome-based pricing in April 2026: Breeze Customer Agent at $0.50 per resolved conversation, Breeze Prospecting Agent at $1.00 per prospect worked. The same announcement cut the Customer Agent rate from $1.00 to $0.50, a deliberate land-and-expand move that prioritises adoption over near-term revenue.

The result is a three-layer model: seats for platform access, credits for AI compute, outcomes for agent value. Each layer maps to a distinct type of AI activity. Structurally, this is the most sophisticated AI commercial architecture in the mid-market segment.

Intercom (Fin AI) charges $0.99 per resolved conversation, entirely separate from seat pricing; resolution means a conversation closed without human escalation. Zendesk introduced value-based resolution tiers in May 2026: simpler interactions charge at a lower rate and higher-value interactions at a higher one, with a base monthly allocation per agent seat before per-resolution billing begins and overages at $1.50 on committed volume or $2.00 pay-as-you-go. Klaviyo has taken the most transparent hybrid approach: a $200 per month base plus $0.70 per conversation beyond the included 50 sessions. Clear base cost, predictable overage, simple to budget.

Salesforce (Agentforce) has built the most architecturally ambitious model in the enterprise market: five distinct entry points, from Foundations, included at no cost on Enterprise Edition and above with enough credits for roughly 10,000 standard agent actions, through Flex Credits at $0.10 per standard action, a flat-rate Conversations option at $2.00 for customer-facing agents under pre-committed contracts, per-user licences, and bundled Editions from $550 per user per month. The underlying logic is coherent, but it is not obvious at first encounter, or second, and choosing between mutually exclusive credit and conversation models at the point of contracting adds real commercial risk.

The flat per-user AI add-on was the default commercial response when AI features launched in 2023 and 2024. Familiar to buyers, simple to forecast, easy to explain. It is also, on this benchmark, the model the market is most actively moving away from.

Microsoft 365 Copilot is the most consequential product still here, at $30 per user per month on top of an existing licence: $360,000 a year for a 1,000-person organisation. The business rate has run at a promotional $18, rising to $21 from July 2026, and the repeated pricing adjustments since launch suggest a position under more pressure than the headline price implies. If Microsoft sustains premium per-user pricing at scale, it strengthens the case for the whole category. If the market keeps moving toward bundling and consumption, Copilot may prove a transitional model rather than the destination.

ClickUp Brain has repriced twice since launch and now runs $9 per user per month for AI Standard and $28 for AI Autopilot, with advanced features consuming credits. Loom prices AI as a plan uplift of roughly $5 to $8 per user per month.

And the others? Slack retired its $10 add-on in August 2025. Notion retired its $8 add-on in May 2025. Monday.com replaced its flat AI fee with a credits model in May 2026. In the space of twelve months, every major productivity tool that launched a flat per-user AI add-on has retired it, repriced it significantly, or moved it into a different commercial model entirely. The standalone add-on is a transitional structure. The destination is tier inclusion or consumption pricing, not a permanent side-car.

Several companies position AI access as the defining differentiator between mid and premium tiers rather than pricing it separately. Slack is the clearest recent migration: the standalone add-on was killed in August 2025 and advanced AI now requires Business+ at $15 per user per month. The move was commercially rational; the add-on was cannibalising tier upgrades. Notion followed the same path and went further, adding a credits layer for its most advanced agentic AI, Custom Agents at $10 per 1,000 credits, on top of the Business-tier requirement, so it now spans two commercial models at once.

Asana has moved to a graduated model: some AI at every paid tier, more capable AI with higher credit ceilings at higher ones. Pipedrive grades AI across all four plans, from a basic sales assistant on the entry tier to forecasting and lead scoring at Premium. Freshdesk gates its autonomous agent capability to Enterprise at $79 to $95 per agent per month, then prices sessions on consumption within it at $100 per 1,000 sessions, with its Copilot agent-assist layer as a separate $29 add-on.

The risk with tier-gating: if lower tiers become materially less competitive than category alternatives because they lack AI, customers switch rather than upgrade. That retention risk typically takes 12 to 18 months to show up in churn data, by which point the damage is done.

The fastest-growing model, and the most consequential structural shift, is the move to credits or token-based pricing: AI capabilities charged against a monthly allotment, with overages priced separately. The approach emerged from API pricing logic and is now permeating B2B SaaS products directly.

GitHub Copilot is making the most significant move. From June 2026, all plans transition to usage-based billing, each including a monthly allotment of AI Credits with additional usage at listed per-token rates; code completions stay outside the meter. The transition has generated significant debate among developers, which captures the broader tension between predictable subscription pricing and usage-based AI consumption.

Figma enforces per-seat credit limits with add-on packs or pay-as-you-go overages at $0.03 per credit. Miro bundles 25 to 50 credits per user per month depending on plan. Canva runs a transparently simple tiered-usage pool across its AI tools. Monday.com moved from its flat add-on to credits at $0.01 per credit; a typical small team currently spends a modest $8 to $12 per month, but the cost now scales directly with adoption. Atlassian (Rovo) has built credits into the tier structure itself, 25, 70 or 150 credits per user per month by plan, so the tier no longer gates AI on or off; it determines how much AI capacity you have.

Snowflake Cortex runs the most granular AI billing architecture in enterprise data, with two parallel billing systems and per-model token rates from roughly $0.12 to $5.10 per million tokens. Its AI Credits documentation was partly prompted by a widely circulated case of a single misconfigured query generating a $5,000 charge: a credible illustration of the exposure token-based billing creates when guardrails are absent.

A group of companies, particularly in accounting and communications software, have deliberately included AI in all plans without separate charges: a retention and positioning play that makes AI a hygiene factor rather than a premium feature. Xero, QuickBooks and Sage have bundled their AI capabilities across tiers. Zoom includes AI Companion in all paid Workplace plans, a calculated bet on retention rather than per-user monetisation across 200 million daily users. Linear includes AI agents even in its free tier, explicit differentiation against Jira in a market where Atlassian charges for AI credits. Grammarly is the limit case: there is no meaningful non-AI version of the product.

The trade-off: bundlers are exchanging potential AI revenue for retention and positioning. The risk is that as AI capabilities become more central and more computationally expensive, bundled pricing becomes structurally unsustainable. The accounting platforms may face a reckoning if AI moves from anomaly detection to full agent-based bookkeeping.

The seat is not being replaced; it is being extended. Only two of the 25+ products have no seat-based component at all. What is changing is what sits on top: credits, consumption charges and outcome fees layered over the seat base to capture AI value a flat rate cannot reach. The companies doing this well have made the layers legible, each mapping to a distinct type of AI activity. The companies doing it badly have created complexity that lands on procurement teams.

The flat per-user add-on is dying faster than expected. Slack, Notion and Monday.com have all retired or replaced standalone AI add-ons within twelve months; ClickUp has repriced twice. The direction is unambiguous.

Credits are proliferating faster than buyers are ready for. Each implementation is slightly different, and collectively they are training enterprise buyers to expect unpredictable AI line items: a problem for budget holders who approved a fixed SaaS spend.

Outcome-based pricing is real, but the definition problem is hard. Resolution-based pricing works for support agents because a resolved ticket is unambiguous. For most AI capabilities no equivalent clean output exists yet, so outcome pricing stays concentrated in autonomous, agentic AI until the measurement problem is solved elsewhere.

Complexity correlates with buyer friction. The products generating the most commercial confusion are generating the most documented complaints about billing unpredictability. The difference is not the number of models; it is whether the architecture is designed for the buyer or for the vendor. HubSpot runs three models simultaneously and stays legible because each maps to a use case. Salesforce’s five options partially overlap and demand org-level architectural decisions at the point of contracting.

Keep the seat, but be deliberate about what you layer on top. The seat is not the problem; treating it as sufficient on its own is. The right question is not whether to replace seats. It is what your AI does that the seat price does not capture, and what the right mechanism is to capture it.

Match the layer to the capability. Agentic AI that delivers measurable outcomes warrants outcome-based pricing on top of the seat. Assistive AI that improves existing workflows suits tier-gating or inclusion. AI with highly variable consumption warrants a credits or consumption layer. The failure mode is picking a mechanism because it is fashionable rather than because it fits the value delivered.

Do not bury the cost. The companies earning buyer goodwill are the ones with transparent, predictable pricing. Opaque credit models and AI bundled invisibly into enterprise tiers are losing the trust dividend even where the underlying AI is strong.

The standalone AI add-on has a short shelf life. Any company running one should be planning the migration path now, toward inclusion where AI becomes core product, or tier-gating where AI defines a tier. A permanent side-car add-on generates noise in sales conversations and typically under-earns what proper tier restructuring would yield.

Do not price for year-one adoption at the expense of year-three monetisation. Several companies are pricing AI cheaply right now, explicitly buying usage. That is rational only with a credible path to commercial monetisation within 18 to 24 months. Otherwise the capability becomes a structural cost without corresponding revenue.

The B2B SaaS AI pricing market is still early, still messy, and still consequential. The seat is not dead, but it is no longer enough on its own. The commercial architectures being established now, what gets layered on top of the seat and how legibly it is structured for the buyer, will shape purchasing expectations for years. Getting that architecture right before competitors lock in the category norm is one of the most important commercial decisions SaaS businesses will make in 2026.

See how these five models play out for a given seat count and AI usage level: try the interactive AI monetisation explorer →