Home / Research / AI Agent Pricing Statistics

Market research · September 22, 2026

AI Agent Pricing Statistics: 2026 Report

This report looks at how commercial AI agent vendors price their products as of 2026. Those prices are the primary list models vendors use - usage, hybrid, per-seat, per-task, and outcome-based fees - across customer support, sales and SDR, coding, IT operations, and back-office agents from vendors such as Salesforce, Intercom, Zendesk, Microsoft, and Sierra.

We tracked public and quoted pricing for 420 agents each quarter from Q3 2025 through Q3 2026, surveyed 600 enterprise buyers, and collected margin data from 150 vendors. We combined this data with figures from ICONIQ, Growth Unhinged, Gartner, and vendor pricing pages.

Share of AI Agent Products by Pricing Model

In the table below, we classify each of the 420 agents by its primary pricing model, meaning the metric that accounts for most of its list price, and compare Q3 2026 with Q3 2025. The last column gives a published price from a well-known vendor using that model.

The Share of AI Agent Products by Pricing Model, 2026

Usage-based pricing leads at 27% of agents, followed by hybrid pricing at 23% and per-seat pricing at 21%. Per-seat pricing lost 13 points in a year, the only model to shrink, and outcome-based pricing absorbed the largest portion of it, gaining 7 points, while usage, hybrid, and per-task models took the rest. Gartner said in July 2026 that agentic AI "breaks the link between user growth and revenue growth for many enterprise software vendors," and our vendor counts show that shift already under way. Vendors that dropped seats most often cited agents doing work that no longer maps to a human login, which makes a per-user fee hard to defend in renewal talks.

Our figures run lower than some survey-based counts because we record only one primary model per agent. ICONIQ found that 23% of software companies used outcome-based pricing in Q2 2026, up from 18% six months earlier, but its respondents blend an average of 1.7 models. Salesforce also illustrates the movement between categories: it launched Agentforce at $2 per conversation and in May 2025 added Flex Credits priced at $0.10 per action, sold in packs of 100,000 credits for $500. Sierra, which charges customers when its agents complete agreed outcomes and does not publish list prices, is one of several private vendors we classified from contract terms that buyers shared with us.

Quarterly Pricing Model Share, Q3 2025-Q3 2026

The table below tracks the share of the same 420 agents on four primary pricing models in each quarter. Per-task and per-conversation pricing, which held at 16% to 17% every quarter, is left out of the quarterly series below.

The Quarterly Pricing Model Share, Q3 2025-Q3 2026

Insights

• We found that outcome-based pricing grew in every quarter we tracked, rising from 5% of agents in Q3 2025 to 12% in Q3 2026. Most of the new outcome-priced agents were customer support products billing per resolution, though a handful of sales agents now bill per qualified meeting booked.
• Our data showed per-seat pricing falling 13 points over the period, with a brief uptick from 30% to 31% in Q1 2026 before its steepest one-quarter drop, 6 points, in Q2 2026. The Q1 uptick came mostly from coding and sales tools that bundled agent features into existing seat plans at annual renewals.
• We observed usage-based pricing overtaking per-seat pricing as the most common primary model in Q2 2026, when it reached 28% against 25% for seats. Usage slipped 1 point to 27% in Q3 2026 as several usage-priced vendors added platform fees and moved into the hybrid group, which rose from 21% to 23% in that quarter.

Typical AI Agent Price Points by Category

Here we break the 420 agents out by category and report the most common primary model, the median list price on that model's unit, and the median annual contract value (ACV) buyers reported for agents in each category.

The Typical AI Agent Price Points by Category, 2026

Customer support is the only category where outcome-based pricing is the most common model, and its median of $1.10 per resolution is between Intercom's $0.99 and Zendesk's $1.50. Coding agents carry the highest median ACV, at $96,000, because seat fees stack with metered usage as engineering teams run more agent sessions, and coding buyers reported that overage charges often exceeded the base seat fee within the first year. Sales and SDR agents have the lowest median ACV, at $41,000, which reflects buyers purchasing only a few $1,150-per-month agent seats at a time. Several SDR buyers told us they cut seat counts at renewal after reply rates fell short of pilot results, which keeps contract values in this category low.

IT operations and back-office agents both price mainly on consumption. The IT/ops median of $0.14 per action runs slightly above the $0.10 per action Salesforce charges through Flex Credits, while back-office vendors meter per document processed, at a median of $0.32. For both categories, unit prices matter less to buyers than volume: the IT/ops median ACV of $54,000 implies roughly 386,000 billed actions a year at the median rate, and buyers in both groups said forecasting that volume was their main difficulty during procurement.

Buyer Pricing Preferences vs. Vendor Pricing Models

In the table below, we compare the model each of our 600 surveyed buyers said they prefer with the share of agents using that model as their primary model in Q3 2026, and we list the budget owner most often named by buyers in each preference group.

The Buyer Pricing Preferences vs. Vendor Pricing Models, 2026

Insights

• We found that 31% of buyers prefer outcome-based pricing, the most of any model, while only 12% of agents use it as their primary model, a 19-point gap. Buyers who prefer it most often said they want vendors to share the risk of an agent that fails to resolve work.
• Our data showed usage and per-seat pricing both oversupplied relative to demand, with vendors using them 9 and 7 points more often than buyers prefer them. Buyers who preferred neither of those models most often cited unpredictable monthly invoices under usage pricing and paying for idle licenses under seat pricing.
• We observed that buyers preferring outcome-based or per-task pricing most often fund agents from customer service or operations budgets, while IT owns the budget in the other three groups, in line with Growth Unhinged's finding that 70% of AI spending still comes from technology or software budgets.

Gross Margin and Inference Cost Pass-Through by Pricing Model

The table below draws on the 150 vendors that shared financial data, grouped by their primary pricing model, and shows average gross margin on agent products, the share passing at least some inference cost through to customers, and the share that repriced in the past 12 months.

The Gross Margin and Inference Cost Pass-Through by Pricing Model, 2026

Outcome-based vendors report an average gross margin of 44%, 17 points below per-seat vendors at 61%, because they absorb the inference cost of every failed or escalated attempt and pass costs through to customers only 22% of the time. Across all 150 vendors, average gross margin is 55%, close to the 53% ICONIQ reported for AI products in 2026. Usage-based vendors pass costs through most often, at 91%, which aligns with ICONIQ's finding that 84% of consumption-priced companies pass at least some of the token bill to customers. Per-seat vendors pass costs through 38% of the time, second lowest after outcome-based vendors, and usually do so through fair-use caps or premium tiers for heavier agent workloads, without an explicit metered charge.

Pricing remains unsettled. Overall, 62% of vendors changed their agent pricing in the past 12 months, led by per-seat vendors at 71%, many of them adding credits or usage tiers. Growth Unhinged found that 75% of B2B software and AI companies changed pricing or packaging over a similar period. Outcome-based vendors repriced least often, at 49%, largely because their contracts already tie revenue to results and leave less room for mid-term changes. Vendors across every model said inference costs, which can vary sharply between simple and complex tasks, remain the main input they watch when setting next year's prices.

Requesting a Copy of This Report

If you would like a PDF copy of this report, including the full category and quarterly data, you can reach out here.

Sources

• AI Agent Pricing Study, AI Industry Reviews, September 2026, New York, New York.
• State of AI: The Builder's Economy, ICONIQ (ICONIQ Analytics), July 2026, San Francisco, California. https://www.iconiq.com/growth/reports/state-of-ai-2026
• The 2026 State of B2B SaaS and AI Monetization Report, Growth Unhinged (Kyle Poyar), May 2026, Boston, Massachusetts. https://www.growthunhinged.com/p/the-state-of-b2b-monetization-in-2026
• Gartner Says $234 Billion in Enterprise Application Software Spend Is at Risk from Agentic Artificial Intelligence, Gartner, July 2026, Stamford, Connecticut. https://www.gartner.com/en/newsroom/press-releases/2026-07-01-gartner-says-us-dollars-234-billion-in-enterprise-application-software-spend-is-at-risk-from-agentic-artificial-intelligence
• Salesforce Introduces New Flexible Agentforce Pricing to Accelerate the Digital Labor Revolution, Salesforce, May 2025, San Francisco, California. https://www.salesforce.com/news/press-releases/2025/05/15/agentforce-flexible-pricing-news/
• Understanding Outcome-Based Pricing: A Results-Driven Framework, Zendesk (Candace Marshall), April 2026, San Francisco, California. https://www.zendesk.com/blog/ai/agentic-ai/outcome-based-pricing/
• Fin Pricing, Intercom, September 2026, San Francisco, California. https://fin.ai/pricing
• Microsoft 365 Copilot Pricing: Copilot Studio, Microsoft, September 2026, Redmond, Washington. https://www.microsoft.com/en-us/microsoft-365-copilot/pricing/copilot-studio