Agent Pricing Tracker: October 2026
Vendors price agents on the unit they can count, and most can count actions, not outcomes. Customer service is the one market where a resolution is countable, so per-outcome pricing lives there, from $0.50 (HubSpot) to $2.00 (Zendesk).
What did we find?
- Credits are the default meter, and the dollar value of a credit is often missing. Salesforce, Microsoft, HubSpot, Lindy, and Clay publish a price per credit. Workday publishes credits per action and no dollar price. Oracle AI units, SAP AI units, and ServiceNow assists are quote only.
- Outcome pricing stays in customer service, and the definitions differ. Zendesk charges only for requests resolved with no escalation. Intercom counts a completed Procedure, including a handoff, as an outcome. Sierra and Decagon publish the model and no rate.
- The seat became the gate, not the price. Salesforce charges $125 per user for the Agentforce add-on, Microsoft $30 for Microsoft 365 Copilot, and Zendesk $50 for Copilot. Usage meters sit on top.
- Vendors that earn on payments include agents in the plan. Ramp, Brex, and BILL put agents inside free or seat plans and earn on interchange and transaction fees. Tabs and Upflow fold agents into a platform fee tiered by customer size.
- Success fees appear only in AR, and no rate is public. Causa Prima takes a success fee and publishes no percentage.
- Model tokens now show through the price. Microsoft bills reasoning tokens directly. Clay meters token-heavy models by usage. Zapier prices AI steps by model tier. Token list prices keep falling, which widens the margin for vendors that charge flat per action.
- Protection against runaway use differs by vendor. Microsoft disables agents at 125% of prepaid capacity, Zapier stops at three times the plan, n8n caps entry plans, and Workday lets agents keep running. Salesforce and HubSpot added hard or default caps in October.
This tracker records how vendors charge for AI agents: the pricing model, the list price, the unit, and what counts as one unit. It covers six markets: AR automation, ERP, AP and spend management, AI sales and marketing tools, agentic payments, and agent platforms and integration. It adds the CRM and customer service vendors that set the benchmarks, and model API token prices as the cost floor. Tags: (vendor claim) marks a price the vendor publishes on its own page or docs, (third-party evidence) a third-party report of a price, and Edition a price carried from one of the market reports in this series, which cite the original source. Every row carries its read date. Currency is USD unless a row says otherwise.
What happened in October 2026. This is the first edition, so it sets the baseline.
- Workday published a full credit rate card and still no dollar price per credit. The 2026-10-12 card adds Decision Intelligence, Extend, and Data Cloud rows and cuts one rate by two thirds. Rising sold agents on credits with no published price (section 4).
- Microsoft’s revised Copilot Studio billing page bills reasoning models by the token. The page, dated 2026-10-27, charges 10 credits per 1,000 tokens on top of the feature rate, with cache reads free (sections 4 and 7).
- AWS moved AgentCore memory to per-GB pricing on 2026-10-06. Short-term memory went from $0.25 per 1,000 events to $1.00 per GB ingested, with new retrieval and storage meters (section 4).
- Hard spend caps shipped at five vendors in one month. Salesforce’s 2026-10-27 Agent Fabric release made hard budget limits generally available, with Flex Credits unchanged at $500 per 100,000. HubSpot, ZoomInfo, Anthropic, and Google shipped caps between 10-08 and 10-23 (sections 4 and 6).
- Workday lets agents run past the credit pool. Console alerts fire at 80, 90, and 100% and nothing deactivates, the opposite of Microsoft’s 125% shutoff (section 6).
- An AR entrant prices on outcomes alone. Causa Prima, funded 2026-10-13, takes a cut of the early payments its agents negotiate (Fortune) (sections 4 and 5).
- Oracle set the ERP pattern on the last day of September. Agentic applications sell as a separate subscription, metered in AI units, with no rate published and availability in the fourth quarter (sections 3 and 5).
1. Summary
Vendors price agents on the unit they can count, and most can count actions, not outcomes. Customer service is the one market where a resolution is countable, so per-outcome pricing lives there, from $0.50 (HubSpot) to $2.00 (Zendesk). Everywhere else the meter is a credit or a task. The seat survives as the gate to the meter. Vendors that earn on payments give agents away.
Seven patterns define agent pricing in October 2026.
- Credits are the default meter, and the dollar value of a credit is often missing. Salesforce, Microsoft, HubSpot, Lindy, and Clay publish a price per credit. Workday publishes credits per action and no dollar price. Oracle AI units, SAP AI units, and ServiceNow assists are quote only.
- Outcome pricing stays in customer service, and the definitions differ. Zendesk charges only for requests resolved with no escalation. Intercom counts a completed Procedure, including a handoff, as an outcome. Sierra and Decagon publish the model and no rate.
- The seat became the gate, not the price. Salesforce charges $125 per user for the Agentforce add-on, Microsoft $30 for Microsoft 365 Copilot, and Zendesk $50 for Copilot. Usage meters sit on top.
- Vendors that earn on payments include agents in the plan. Ramp, Brex, and BILL put agents inside free or seat plans and earn on interchange and transaction fees. Tabs and Upflow fold agents into a platform fee tiered by customer size.
- Success fees appear only in AR, and no rate is public. Causa Prima takes a success fee and publishes no percentage.
- Model tokens now show through the price. Microsoft bills reasoning tokens directly. Clay meters token-heavy models by usage. Zapier prices AI steps by model tier. Token list prices keep falling, which widens the margin for vendors that charge flat per action.
- Protection against runaway use differs by vendor. Microsoft disables agents at 125% of prepaid capacity, Zapier stops at three times the plan, n8n caps entry plans, and Workday lets agents keep running. Salesforce and HubSpot added hard or default caps in October.
2. Pricing models defined
| Model | How it works | Examples in section 3 | Who carries the risk |
|---|---|---|---|
| Seat | A fixed fee per named user per month, agent features included | Microsoft 365 Copilot, Odoo Custom, Brex Premium | Vendor carries usage cost. Buyer pays for idle seats and gains nothing as agents replace users |
| Seat plus agent add-on | A base seat plus a per-user agent license | Salesforce Agentforce add-ons, Zendesk Copilot | Buyer pays twice per user. Vendor keeps seat revenue as headcount falls only while the add-on holds |
| Credits and consumption | A prepaid or metered pool, drawn down at published rates per action, token, or minute | Salesforce Flex Credits, Copilot Studio, Workday Flex Credits, HubSpot Credits, Clay, Lindy | Buyer carries volume and design risk. A verbose agent costs more. Expiry moves unused value back to the vendor |
| Per action | A fixed charge per step, task, run, or tool call | Zapier tasks, n8n executions, Salesforce $0.10 per action | Buyer pays for every step, useful or not. Vendor carries model cost inside the flat rate |
| Per outcome or resolution | A charge only when the agent resolves the request, by a vendor-set definition | Intercom Fin, Zendesk automated resolutions, HubSpot Customer Agent, Sierra | Vendor carries failure and model cost. Buyer carries definition risk, since the vendor decides what counts |
| Gain-share | A share of measured savings or recovered cash against agreed criteria | Causa Prima success fee | Vendor carries delivery risk. Buyer carries measurement disputes and pays more when results are good |
| Platform fee plus usage | A fixed platform or infrastructure charge plus metered compute or calls | Tabs Launch, Upflow Payments, AWS AgentCore, Claude Managed Agents | Shared. Buyer commits to the floor, and usage above it scales with traffic |
| Free with payments revenue | Agents included at no charge, with revenue earned on card interchange or payment fees | Ramp, Brex, Stripe agentic commerce, Visa Intelligent Commerce | Buyer pays through payment flow. The agent cost is hidden in spend volume and switching cost |
3. The tracker
Each row is one vendor product. “Quote only” means the vendor publishes no list price and gates price behind sales. Rows tagged Edition carry the read date of the market report they come from.
| Vendor | Market | Product | Model | List price | Unit | What counts as the unit | Source | Read | Tag |
|---|---|---|---|---|---|---|---|---|---|
| Salesforce | CRM and service | Agentforce Flex Credits | Credits | $500 per 100,000 credits (EUR 500, GBP 400). 20 credits ($0.10) per action, 30 ($0.15) per voice action. Overage billed monthly in arrears | Action | One agent action, drawn from the org credit pool. Data 360 credits are extra | Salesforce | 2026-10-30 | (vendor claim) |
| Salesforce | CRM and service | Agentforce conversations | Per conversation | $2 per conversation (GBP 1.60). $2 per Help Agent resolution, per the AI sales edition | Conversation, resolution | One customer conversation with an agent, or one resolution by the Help Agent | Salesforce | 2026-10-30 | (vendor claim) |
| Salesforce | Agent platforms | Agentforce Foundations | Free tier | $0 per org | Org | Entry access. Agent use above it draws Flex Credits | Agent Platforms and Integration Market report | 2026-10-30 (platforms edition) | Edition |
| Salesforce | CRM and service | Agentforce add-ons (Sales, Service, Field Service) | Seat plus agent add-on | $125 per user per month. Industries add-ons $150 | User per month | Named user | Salesforce | 2026-10-30 | (vendor claim) |
| Salesforce | CRM and service | Agentforce User License and Coworker License | Seat plus credits | $5 per user per month plus Flex Credits. Coworker License $50 | User per month | Named user, with agent usage drawn from credits | Salesforce | 2026-10-30 | (vendor claim) |
| Salesforce | CRM and service | Max Edition | Seat with bundled credits | From $550 per user per month, including the Agentforce add-on and 2.75M Flex Credits per org per year. Sales Core $195 and Advanced $395 per user per month | User per month | Named user | Salesforce, Sales pricing (read by the AI sales edition) | 2026-10-30 | (vendor claim) |
| HubSpot | CRM and service | HubSpot Credits | Credits | $0.010 per credit ($10 per 1,000) on the pricing page. $9 per 1,000 on annual commitment, per a pricing tracker | Credit | Breeze AI and agent usage, at rates in HubSpot’s rate sheet | HubSpot, PricingSaaS | 2026-10-30 | (vendor claim), (third-party evidence) for $9 |
| HubSpot | CRM and service | Breeze Customer Agent | Per outcome, through credits | 50 credits (about $0.50 at $0.010 a credit) per resolved conversation since 2026-04-14. Before that, 100 credits per conversation, resolved or not | Resolved conversation | A resolved chat. Unresolved chats cost nothing | HubSpot catalog, HubSpot for the start date (via the AI sales edition) | 2026-10-30 | (vendor claim) |
| HubSpot | Sales and marketing | Marketing Hub Professional | Seat plus included credits | $800 per month with 3 core seats and 3,000 credits a month. More seats from $45. $3,000 onboarding | Month | Plan, with credits reset monthly | HubSpot | 2026-10-30 | (vendor claim) |
| HubSpot | Sales and marketing | Prospecting, Content, and Nurture agents | Per unit of work, through credits | Prospecting 100 credits (about $1) per lead recommended, outcome-priced since 2026-04-14. Content 1,000 credits per piece. Nurture 10 credits per email | Lead, content piece, email | A lead recommended for outreach, not a reply or a meeting. Before April, a monthly charge per enrolled contact | HubSpot catalog, HubSpot | 2026-10-30 (AI sales edition) | (vendor claim) |
| Intercom | CRM and service | Fin AI Agent | Per outcome | $0.99 per outcome. A monthly minimum applies when Fin runs on another helpdesk (the page’s example is 50 outcomes) | Outcome | The customer confirms resolution, asks nothing more after Fin answers, or Fin completes a Procedure, handoffs included. Charged once per conversation | Intercom | 2026-10-30 | (vendor claim) |
| Intercom | CRM and service | Helpdesk seats | Seat | Essential $29 ($19 for new customers), Advanced $85, Expert $132 per seat per month. Fin alone needs no seat | Seat per month | Human agent seat | Intercom | 2026-10-30 | (vendor claim) |
| Zendesk | CRM and service | AI agents | Per outcome | $2.00 per automated resolution pay-as-you-go, $1.50 committed. 5 to 10 included per agent per month by plan | Automated resolution | A request the AI agent resolves with no escalation to a human | Zendesk | 2026-10-30 | (vendor claim) |
| Zendesk | CRM and service | Copilot | Seat plus agent add-on | $50 per agent per month paid yearly, on Suite Professional ($115 yearly, $149 monthly) and above | Agent seat per month | Human agent seat. Professional includes 5 Copilot uses per agent, capped at 500 per instance | Zendesk | 2026-10-30 | (vendor claim) |
| Sierra | CRM and service | Sierra agents | Per outcome, consumption for routing | Quote only. Third-party estimates run $1 to $2.50 per resolution and platform contracts from about $150,000 a year, unconfirmed by Sierra | Agreed outcome | A resolved conversation, saved cancellation, upsell, or cross-sell, with criteria agreed upfront. Unresolved and escalated work “in most cases” free | Sierra, CloudTalk | 2026-10-30 | (vendor claim) model, (third-party evidence) price |
| Decagon | CRM and service | Decagon agents | Per conversation or per resolution | Quote only. Third-party estimates put per-conversation at about $0.99 and a platform fee at about $50,000 a year, both unconfirmed | Conversation, or resolved conversation | Per-resolution is the higher rate with escalations free. Decagon says most customers pick per conversation | Decagon, Featurebase | 2026-10-30 | (vendor claim) model, (third-party evidence) price |
| Microsoft | Agent platforms | Copilot Studio credits | Credits | $200 per month for a pack of 25,000 Copilot Credits (CAD 271.40, GBP 153.80), or $0.008 a credit. Pay-as-you-go $0.01 per credit, billed through Azure in arrears at month end. Up to 20% off with a one-year prepurchase plan of commit units | Copilot Credit | Tenant-wide pool. Unused pack credits do not roll over | Microsoft, Pay-as-you-go meters | 2026-10-30 (meters page dated 2026-10-14) | (vendor claim) |
| Microsoft | Agent platforms | Copilot Studio feature rates | Per action, through credits | Classic answer 1 credit, generative answer 2, agent action 5, tenant graph grounding 10, 100 agent flow actions 13 | Feature event | One interaction can draw several meters. Free when the user holds a Microsoft 365 Copilot license and the agent acts as that user | Microsoft Learn | 2026-10-30 (page dated 2026-10-27) | (vendor claim) |
| Microsoft | Agent platforms | Copilot Studio reasoning and voice | Tokens, through credits | Reasoning models 10 credits per 1,000 tokens on top of the feature rate. Voice 10, 35, or 75 credits a minute by tier | Token, minute | Input and output tokens including reasoning tokens. Cache reads free | Microsoft Learn | 2026-10-30 | (vendor claim) |
| Microsoft | ERP | Dynamics 365 ERP MCP server | Per action, through credits | 0.1 Copilot Credits per tool call from outside clients. Inside Copilot Studio it bills as an agent action. Finance Premium and SCM Premium users are exempt outside Copilot Studio | Tool call | One create, read, update, or delete call. Agent identities need no user license | Microsoft Learn | 2026-10-30 | (vendor claim) |
| Microsoft | CRM and service | Microsoft 365 Copilot | Seat | $30 per user per month billed yearly, $31.50 paid monthly on an annual commitment. Needs a qualifying Microsoft 365 plan | User per month | Named user | Microsoft | 2026-10-30 | (vendor claim) |
| Microsoft | CRM and service | Microsoft 365 Copilot Business | Seat | $21 per user per month annual, $18 as a first-year promotion to 2026-12-31, $25.20 monthly | User per month | Named user, for existing eligible Business plan customers | Microsoft | 2026-10-30 | (vendor claim) |
| Microsoft | Agent platforms | Agent 365 and Microsoft 365 E7 | Seat | Agent 365 $15 per user per month, paid yearly. E7 $99 includes Agent 365 (E7 without Teams $90.45), annual commitment | User per month | Named user. Agent 365 governs agents across the tenant | Microsoft | 2026-10-30 | (vendor claim) |
| Microsoft | AP and spend | Business Central Payables Agent | Per action, through credits | No rate published. Billed in Copilot Credits “based on interaction complexity” | Copilot Credit | Agent interactions. The agent drafts invoices for a supervisor and does not post | Microsoft Learn | 2026-10-30 (AP edition) | (vendor claim) |
| Microsoft | Sales and marketing | Dynamics 365 Sales Premium | Seat with bundled credits | $150 per user per month with 1,000 Copilot Credits | User per month | Named user. No separate per-credit rate on the page | Microsoft | 2026-10-30 (AI sales edition) | (vendor claim) |
| ServiceNow | CRM and service | Foundation, Advanced, and Prime tiers with assists | Seat plus pooled consumption | Quote only. Benchmarked deals run about $70 to $100 per user (Foundation) and $160 to $200 or more (Prime) before consumption. A rate card of $0.20 per assist, documented by Raymond James in June 2026. Negotiated top-ups $0.02 to $0.10 per assist | Assist | Tenant-level pool. Multi-step agent actions draw more assists than a summary. The ITSM AI Starter Pack carries 150,000 assists for 25 users. Tier allocations unpublished | Redress tiers, Redress assists | 2026-10-30 (articles updated 2026-08-16 and 2026-08-28) | (third-party evidence) |
| Workday | ERP | Flex Credits | Credits | No dollar price per credit published, confirmed on Workday’s page and in two buyer-side guides. Credits per action run from 1 (information retrieval) to 750 (Talent Rediscovery per requisition) | Credit | One agent action or skill run, plus API calls above an annual allowance at 60 credits per 10,000 calls. Annual purchase with complimentary credits by company size | Workday, UsagePricing, Redress | 2026-10-30 (card dated 2026-10-12) | (vendor claim) no price, (third-party evidence) rates |
| Workday | ERP | Finance agents (Contract Negotiation, Revenue Contract, Financial Audit) | Per action, through credits | 500 credits per document redlined, 25 per revenue contract, 60 per audit sample | Document, contract, sample | One unit of the named work | UsagePricing | 2026-10-30 | (third-party evidence) |
| Oracle | ERP | Fusion Agentic Applications (Fusion Claw) | Platform subscription plus consumption | US$500,000 a year per Agentic Applications service (“Each”, minimum 1). AI units US$1,000 per 100,000 (US$0.01 a unit). Fusion AI Agents US$150 per named user a month (minimum 50) or US$5 per employee a month (minimum 500). Per-action unit table and included allowances not read | Year, AI unit, user, or employee | One agentic application service, or pooled AI units consumed by actions | Oracle global price list, dated 2026-10-08, read through the ERP edition | 2026-10-30 | (vendor claim) |
| Oracle | ERP | NetSuite AI Units | Seat with pooled units | 1,000 AI Units per user per month in a shared pool. No currency price | AI unit | About 10 units for a simple question and 50 to 200 for a research query. Bill Capture uses none | Boston SD | 2026-10-30 (AP edition) | (third-party evidence) |
| SAP | ERP | Joule agents | Credits (AI units) | Standard agent action 0.02 AI units, advanced 0.05. €7 per AI unit sold in blocks of 100, so about €0.14 per standard action and €0.35 per advanced. The €7 comes from an SAP user-group deck, not a price list | Agent action | One agent action. Units sit in an annual pool that expires after 12 months | SAP pricing page, SAP user-group deck, 2026, read through the ERP edition | 2026-10-30 | (vendor claim) units, (vendor claim) deck price |
| SAP | ERP | Joule Studio and Enterprise AI Gateway | Free promotion, then unpriced premium | Custom-agent runtime free through December 2026 for selected customers. Design-time tools free “within fair-use limits”. Enterprise AI Gateway for outside agents is a premium offering due Q1 2027, unpriced | Agent runtime, gateway call | Runtime of custom agents. Outside agents reach SAP data through the metered gateway | EuropaWire copy of SAP’s release | 2026-10-30 (ERP edition, release 2026-10-27) | (vendor claim) |
| Odoo | ERP | Custom plan | Seat, agents included | Prices conflict on reads the same day. This tracker read US$61.00 per user per month (US$76.20 after the first-year discount) and Standard US$31.10. The ERP edition read Custom at US$49 to 76.20 and Standard at US$24.90 to 38.90. Standard has no agents | User per month | Named user. Odoo 20 says “All AI features now need credits, which can be bought via IAP”, at unpublished rates | Odoo, Odoo 20 | 2026-10-30 | (vendor claim) |
| IFS | ERP | IFS Intelligence | Bundled under fair use | No price. “a universal enterprise-wide license” plus AI Executions, “not a credit model but a fair usage cap” | AI Execution | Executions inside an allotment | IFS | 2026-10-30 (ERP edition, release 2026-10-13) | (vendor claim) |
| Infor | ERP | Velocity Suite | Bundled under fair use | No price. “unlimited access within fair business use” | Not metered | Fair business use | ERP Market report | 2026-10-30 (ERP edition) | Edition |
| Unit4 | ERP | AI for your world | Free promotion | AI with no commitment to 2027-08-31 for customers who register by 2026-12-31 | Not metered during the offer | Registered customers | Unit4 | 2026-10-30 (ERP edition) | (vendor claim) |
| Acumatica | ERP | AI units in 2026 R2 | Credits | Unpriced. Threshold warnings as the pool runs down | AI unit | Agent and AI feature use | ERP Today | 2026-10-30 (ERP edition) | (third-party evidence) |
| Intuit | ERP | Agentic Books Close | Add-on fee | No rate. “additional fees may apply”. The product is “coming soon” | Not published | Not published | Intuit | 2026-10-30 (ERP edition, release 2026-10-27) | (vendor claim) |
| Digits | ERP | Accounting firm pricing | Per outcome | Rate unpublished. Firms pay only “when the work is done”. A third-party profile puts the trigger at 95% or more of a client’s transactions handled touch-free | Client handled touch-free | Share of a client’s transactions with no human touch | Digits blog via ERP Market report, Software Connect | 2026-10-30 (ERP edition) | Edition, (third-party evidence) for the trigger |
| Sidetrade | AR automation | Aimie agents under SAFE | Subscription with per-agent fees | Six figures, 46-month initial terms, per-agent fees. Task volume and token equivalents bundled, compute price guaranteed through multi-year contracts | Agent, task | Agent count and bundled task volume | GlobeNewswire | 2026-10-18 | (vendor claim) |
| BlackLine | AR automation | Agents | Moving from tokens to outcomes | No price published. Management declared a move from token-based to outcome-based agent pricing | Outcome (stated intent) | Not defined publicly | Transcript | 2026-10-18 | (vendor claim) |
| Tabs | AR automation | Launch plan | Platform fee, agents included | $2,000 per month for up to $5M revenue and 100 active contracts. Growth, Scale, and Enterprise quote only | Month, sized by active contracts | Contract, Collections, Billing, and Revenue agents included in every plan | Tabs | 2026-10-30 | (vendor claim) |
| Upflow | AR automation | Collections, AP Portal, and Cash Application agents | Tiered platform fee, agents included | Quote only. Tiers sized by annual invoiced revenue, from Starter (to $10M) to Enterprise (above $100M) | Annual invoiced revenue | Collection and AP Portal agents from Scale, Cash Application from Accelerate, under fair usage | Upflow | 2026-10-30 | (vendor claim) |
| Upflow | AR automation | Upflow Payments | Platform fee plus take-rate | $390 per month plus 3.5% + $0.30 per card payment and 0.8% per ACH debit, capped at $5 | Payment | One collected payment | Upflow | 2026-10-30 | (vendor claim) |
| Causa Prima | AR automation | Negotiation agents | Success fee | Rate unpublished. A cut of the early payments its agents negotiate | Negotiated early payment | Agent-to-agent agreement on payment terms | Fortune | 2026-10-18 | (third-party evidence) |
| Ramp | AP and spend | Ramp platform and agents | Free with payments revenue | Free plan $0 per user. Plus $15 per user per month plus a team-size platform fee. Agents carry no separate price, and Agent Cards (cards issued to AI agents) carry no stated price per the payments edition. The AR and procurement add-ons are unpriced | User per month | Named user | Ramp | 2026-10-30 | (vendor claim) |
| BILL | AP and spend | AP and AR plans with W-9 and Invoice Coding agents | Seat plus transaction fees, agents included | Essentials $49, Team $65, Corporate $89 per user per month. ACH $0.59, card acceptance 2.9% | User per month, payment | W-9 Agent in all AP plans. Invoice Coding Agent from Team up. Transaction Agent included “at no extra cost” for Spend & Expense customers | BILL, product updates | 2026-10-30 | (vendor claim) |
| Zip | AP and spend | AI Spend Automation with Superagents | Credits plus services | No price. Consumption credits bundled with forward-deployed engineers | Credit | Not published | Zip | 2026-10-30 (AP edition, post 2026-06-02) | (vendor claim) |
| Brex | AP and spend | Brex AI expense assistant | Free with payments revenue | Essentials $0, Premium $12 per user per month. AI features carry no separate price | User per month | Named user | Brex | 2026-10-30 | (vendor claim) |
| Tipalti | AP and spend | Tipalti AI agents | Platform fee plus transaction fees | Accounts Payable from $99 per month, Mass Payments from $249. Agents carry no separate price. Per-payment fees quote only | Month, invoice, payment | Plan plus per-invoice and per-payment charges | Tipalti | 2026-10-30 | (vendor claim) |
| Clay | Sales and marketing | Launch and Growth plans | Credits on two meters | Launch $167 per month and Growth $446 per month, billed annually. Data credits about $0.05 each at entry, actions about $0.004 | Data credit, action | Data credits buy marketplace data. Actions pay for runs, enrichment, AI calls, and exports. Empty enrichments cost nothing | Clay | 2026-10-30 | (vendor claim) |
| Clay | Sales and marketing | Claygent and AI models | Fixed or token-based credits | About 80% of models cost a fixed number of data credits per task. Token-heavy models (GPT-5.1, Claude 4.6 Sonnet) bill by tokens used | Task, token | One AI task. Bring-your-own keys skip data credits but each run still counts as an action | Clay | 2026-10-30 | (vendor claim) |
| 11x | Sales and marketing | Alice (SDR) and Julian (voice and chat) | Subscription by volume | Alice Growth from $3,750 a month, billed annually (5 users, 2,000 new prospects a month). Julian Growth from $5,333 a month for voice (2,000 calls) or $2,417 for chat (3,000 chats) | Month | Prospects, calls, or chats in the plan | Alice, Julian | 2026-10-30 | (vendor claim) |
| Artisan | Sales and marketing | Ava (AI BDR) | Plan by volume | Not published. Plans scale by leads (about 2,500 or 6,000 a month), mailboxes, and dialer seats, and the page mentions no credits | Plan | Leads, mailboxes, and dialer seats | Artisan | 2026-10-30 | (vendor claim) model only |
| AiSDR | Sales and marketing | AiSDR | Per unit of work | $1.50 per person prospected on pay as you go. Warmed mailbox $33, managed campaign $149, fully managed $2,500 a month | Person prospected | One prospect the agent researches and contacts | AiSDR | 2026-10-30 | (vendor claim) |
| Regie.ai | Sales and marketing | RegieGO Pro and Team | Seat plus credits | Pro $41 to $49 a month. Team $125 to $150 per seat a month | Credit | 5,000 (Pro) and 15,000 (Team) credits per seat a month | Regie.ai | 2026-10-30 (AI sales edition) | (vendor claim) |
| ZoomInfo | Sales and marketing | GTM.AI pay as you go | Credits without seats | $0.38 per data credit at the $1,000 tier, $0.05 per AI credit. “no seat fees” | Data credit, AI credit | Credit expiry conflicts on the page (“expire 12 months after they are granted” and “never expire while your account is active”) | GTM.AI | 2026-10-30 (AI sales edition) | (vendor claim) |
| Common Room | Sales and marketing | Essential | Seat plus credits | $2,500 a month, billed annually | Month | 5 seats, 5,000 research credits, and 2,500 Prospector credits | Common Room | 2026-10-30 (AI sales edition) | (vendor claim) |
| Unify | Sales and marketing | Base and Pro | Seat plus credits | $20 and $60 per seat a month | Seat per month | 800 and 2,400 credits per seat a month | Unify | 2026-10-30 (AI sales edition) | (vendor claim) |
| Amplemarket | Sales and marketing | Startup | Seat plus credits | $600 a month, billed annually | Month | 2 users plus per-user credits | Amplemarket | 2026-10-30 (AI sales edition) | (vendor claim) |
| Qualified | Sales and marketing | Piper (AI SDR) | Plan by volume | Not published | Plan | Web sessions plus active email contacts | Qualified | 2026-10-30 (AI sales edition) | (vendor claim) model only |
| Jasper | Sales and marketing | Pro | Seat plus credits | $69 per seat monthly, $59 annual | Seat per month | Agents run on credits at a rate card the AI sales edition did not read | Jasper | 2026-10-30 (AI sales edition) | (vendor claim) |
| Writer | Sales and marketing | Palmyra X6 | Tokens | $2 input and $8 output per million tokens. MCP access included in existing seats from 2026-10-27, per the AI sales edition | Million tokens | Input and output tokens | Writer | 2026-10-30 (AI sales edition) | (vendor claim) |
| Klaviyo | Sales and marketing | Composer on the free tier | Dollars of AI usage | ”$5 of Composer usage/month” on the free tier | Dollar of usage | AI usage metered in dollars, not credits | Klaviyo | 2026-10-30 (AI sales edition) | (vendor claim) |
| Apollo | Sales and marketing | Free, Basic, Professional, Organization, Custom | Seat plus credits | Prices not on the readable page. Third parties read Basic $49, Professional $79, and Organization $119 per seat per month billed annually (Organization three-seat minimum). Monthly billing $60, $94, and $149 per one guide. Annual credits per seat 30,000, 48,000, and 72,000 in one guide, far lower in others | Credit | Emails 1 credit, phone numbers 8, plus mailboxes and AI-researched data. Credits pooled across the team, granted at cycle start, unused credits expire each cycle | Apollo, Zeliq (read Apollo’s page 2026-07-27), Docket (verified April 2026) | 2026-10-30 | (vendor claim) model, (third-party evidence) prices |
| Stripe | Agentic payments | Shared Payment Tokens | Per token plus card take-rate | $0.15 per token issued, on top of US card processing at 2.9% + 30¢ | Shared Payment Token | A scoped card credential an agent uses to pay. Card minimum $0.50 | Stripe, docs | 2026-10-30 | (vendor claim) |
| Stripe | Agentic payments | Machine payments (MPP, x402) | Take-rate on payments | Fee not stated on the docs page. Settlement minimum 0.01 USDC, with sub-cent charges inside MPP sessions | Payment | An agent paying for an API call or service | Stripe docs | 2026-10-30 | (vendor claim) |
| Stripe | Agentic payments | Issuing for agent cards | Per card plus FX | $0.10 per virtual card. Cross-border 1% + 30¢, plus 1% for conversion | Card, transaction | One card issued to an agent or one spend | Stripe | 2026-10-30 | (vendor claim) |
| Stripe | Agentic payments | Stablecoin payments and payouts | Take-rate on payments | Through Link 0.8%, a promotional rate through 2027-01-01, then “+0.2%” as the page states it. Global 1.5% including fiat conversion, screening, and gas. Payouts $0.50 + 0.5% | Payment, payout | One successful stablecoin payment or payout | Stripe | 2026-10-30 (payments edition) | (vendor claim) |
| Stripe | Agentic payments | Treasury for agents | Not stated | No price published | n/a | Business accounts agents can move money from, with human approval rules | Stripe | 2026-10-30 (payments edition, launch 2026-10-21) | (vendor claim) |
| x402 Foundation | Agentic payments | x402 protocol | Free protocol | $0 protocol fee. “nominal payment network fees” apply | Transaction | Fees fall to the rail and the facilitator (see Coinbase) | x402.org | 2026-10-30 (payments edition) | (vendor claim) |
| Open Standard | Agentic payments | OUSD stablecoin on Stripe | Per transaction | Dollar to OUSD conversion free on Stripe. The issuer charges “a small, predictable transaction fee”, amount not stated | Transaction | One OUSD payment | Stripe | 2026-10-30 (payments edition, post 2026-09-30) | (vendor claim) |
| Meta | Agentic payments | Muse shopping agent | Share of commerce | Free for most users. Meta plans to keep “a percentage of the commerce”, rate not stated. A $20 or $100 monthly tier appears only in a secondary summary | Purchase | Commerce routed through the agent | TechCrunch | 2026-10-30 (payments edition, article 2026-10-21) | (vendor claim) via (third-party evidence) |
| Mastercard, American Express, Adyen | Agentic payments | Agent Pay, Agent Pay for Machines, ACE Developer Kit, and Adyen Agentic | Free with payments revenue (inferred) | No fee disclosed | Not published | Not published | Agentic Payments Market report | 2026-10-30 (payments edition) | Edition |
| Perplexity | Agentic payments | Shopping with PayPal checkout | Free | Free for US users | n/a | Agent shopping and checkout through PayPal | Perplexity | 2026-10-30 (payments edition, post 2025-11-25) | (vendor claim) |
| OpenAI, Google | Agentic payments | ACP checkout, UCP, and AP2 | Not disclosed | No merchant or agent fee disclosed. A 4% ChatGPT merchant fee circulates on low-authority blogs, unconfirmed | Not published | Not published | Agentic Payments Market report | 2026-10-30 (payments edition) | Edition |
| Coinbase | Agentic payments | CDP x402 Facilitator | Per settled transaction | First 1,000 onchain transactions a month free, then $0.001 each. Verification free | Onchain settlement | One settled payment on Base, Solana, Polygon, Arbitrum, and other listed networks | Coinbase | 2026-10-30 | (vendor claim) |
| Visa | Agentic payments | Intelligent Commerce and Trusted Agent Protocol | Free with payments revenue (inferred) | No fee published. Commercial terms come from regional product teams | Not published | Not published | Visa | 2026-10-30 | (vendor claim) |
| Zapier | Agent platforms | Pro and Team plans | Per action (task) | Pro from $19.99 per month for 750 tasks, Team from $69 for 2,000, both annual. Overage 1.25 times base (annual) or 2.5 times (monthly), hard stop at 3 times the plan | Task | A successful step. An MCP tool call is 2 tasks. Failed actions, triggers, and built-in tools are free. AI steps priced by model tier | Zapier | 2026-10-30 | (vendor claim) |
| Zapier | Agent platforms | Zapier Agents | Per action (activity) | Free 400 activities a month. Pro 1,500 activities a month from about $33.33 per month annual | Activity | One agent step, capped at 40 per run on paid plans. Separate from the task pool | Zapier | 2026-10-30 | (vendor claim) |
| n8n | Agent platforms | Cloud Starter and Pro | Per run (execution) | €20 per month for 2,500 executions, €50 for 10,000, billed annually. Agents, MCP, and AI nodes included | Execution | One full workflow run, whatever its step count or data volume | n8n | 2026-10-30 | (vendor claim) |
| n8n | Agent platforms | Business (self-hosted) and Community | Per run, self-hosted | Business €667 per month for 40,000 executions. Community edition free to self-host with your own model key | Execution | One full workflow run | n8n | 2026-10-30 | (vendor claim) |
| Workato | Agent platforms | Workato ONE with Agent Studio and Genies | Quote only | Quote only. Agent Studio ships only in Workato ONE. Pre-Packaged Genies are an unpriced add-on | Task | Task consumption by capability | Workato | 2026-10-30 | (vendor claim) |
| Lindy | Agent platforms | Team plan | Credits | $49.99 per month for 5,000 credits. Top-ups $10 per 1,000 | Credit | 2 to 250 credits per everyday ask, 250 to 1,000 per deep task, up to 2,500 per large build | Lindy | 2026-10-30 | (vendor claim) |
| Make | Agent platforms | Core, Pro, and Teams | Credits | Prices not on the readable page. Free plan 1,000 credits a month. A third party read Core $10.59, Pro $18.82, and Teams $34.12 a month at 10,000 credits, or $9, $16, and $29 a month billed annually. Other guides show higher figures | Credit | Most actions cost 1 credit, AI features more at unpublished rates. External model calls bill separately from the model provider. Credits expire at term end | Make, Jetadmin (read Make’s page 2026-09-26) | 2026-10-30 | (vendor claim) model, (third-party evidence) prices |
| UiPath | Agent platforms | Agentic automation plans | Seat plus quote | Basic from $25 per month with limited agent calls. Standard and Enterprise quote only | Month, user, robot | Users and robots, with agent calls limited on Basic | UiPath | 2026-10-30 | (vendor claim) |
| AWS | Agent platforms | Bedrock AgentCore | Platform fee plus usage | Runtime $0.0895 per vCPU-hour and $0.00945 per GB-hour (v1), $0.1276 and $0.0169 on demand (v2, the rate the platforms edition cites). Gateway $0.005 per 1,000 calls. Policy $0.000025 per authorization | vCPU-hour, call | Compute while the agent runs, each tool call, each policy decision. Model tokens extra | AWS | 2026-10-30 | (vendor claim) |
| Anthropic | Agent platforms | Claude Managed Agents | Tokens plus runtime | $0.08 per session-hour plus model tokens. Web search $10 per 1,000 searches | Session-hour, token | Time in running status, plus tokens at model rates | Anthropic | 2026-10-30 | (vendor claim) |
| Anthropic | Agent platforms | Claude Enterprise | Seat plus usage | $20 per seat plus API usage, 20-seat minimum | Seat, token | Named user, with usage billed at API rates | Claude pricing | 2026-10-30 (platforms edition) | (vendor claim) |
| LangChain | Agent platforms | LangSmith deployments | Runtime consumption | 0.0675 LSU per vCPU-hour. No dollar price per LSU in the platforms edition | vCPU-hour | Deployed agent compute | Agent Platforms and Integration Market report | 2026-10-30 (platforms edition) | Edition |
| Agent platforms | Gemini Enterprise | Seat plus consumption | Google’s pricing page not readable. Third-party trackers report Business $21, Standard $30 annual or $35 monthly, and Plus $50 annual or $60 monthly per user per month. A UK G-Cloud price document lists Business at $25 and Standard at $35. Agent Platform compute $0.085 per vCPU-hour and storage $0.30 per GiB-month on top | User per month, vCPU-hour | Named user, with agent compute, sessions, and memory metered separately. Sessions and Memory Bank began billing 2026-09-01. Google told Futurum its coworker agent carries no per-seat license, and a $30 figure attributed to Futurum is unconfirmed (platforms edition) | Ecorpit (2026-07-28, citing Coworker AI and GoSearch), G-Cloud document (2026-01-30, search snippet) | 2026-10-30 | (third-party evidence) | |
| Anthropic | Model API | Claude models | Tokens | $0.10 to $10 input and $0.50 to $50 output per million tokens across current models (section 7) | Million tokens | Input, output, and cache tokens | Anthropic | 2026-10-30 | (vendor claim) |
| OpenAI | Model API | GPT models | Tokens | $0.10 to $10 input and $0.50 to $50 output per million tokens across current GPT-6 models (section 7) | Million tokens | Input, cached input, and output tokens | OpenAI | 2026-10-30 | (vendor claim) |
| Model API | Gemini models | Tokens | $0.25 to $4 input and $1.50 to $18 output per million tokens across current Gemini 3 models (section 7) | Million tokens | Input and output tokens, thinking included | 2026-10-30 | (vendor claim) |
Row count: 94.
4. Changes in October 2026
Every row is dated to the change or the first public record of it. The first row comes from 30 September and frames the quarter.
| Date | Vendor | Change | What it means for the buyer | Source | Tag |
|---|---|---|---|---|---|
| 2026-09-30 | Oracle | Agentic applications on Fusion Claw sell as a separate subscription, metered in AI units, available in the fourth quarter | ERP agents carry their own line item, and no public rate exists to budget against | TechTarget | (third-party evidence) |
| 2026-10-06 | AWS | AgentCore short-term memory moved from $0.25 per 1,000 events to $1.00 per GB ingested, with $0.20 per GB retrieval and $0.10 per GB-month storage | Cost now tracks payload size, not event count. Agents with large memories pay more | AWS | (vendor claim) |
| 2026-10-07 | Anthropic | Sonnet 5.5 cache reads cut from $0.20 to $0.10 per million tokens | Repeated prompts get cheaper while credit prices built on them stay put | Anthropic, via the platforms edition | (vendor claim) |
| 2026-10-08 | Per-project spend caps in the Gemini agent | The buyer sets the ceiling, and the vendor’s invoice no longer carries the overrun | Agent Platforms and Integration Market report | Edition | |
| 2026-10-09 | Oracle | The Register reports per-user, per-employee, and consumption options for agentic applications | The first sign of how Oracle will meter Fusion Claw, still with no rate | The Register, via the ERP edition | (third-party evidence) |
| October 2026 | AWS | Runtime committed-baseline rates of $0.0997 per vCPU-hour and $0.0132 per GB-hour (v2) against $0.1276 and $0.0169 on demand, listed as launching by October | About 22% off v2 compute for a commitment, by arithmetic from the listed rates | AWS | (vendor claim) |
| 2026-10-08 | Workday | Credit policy added Marketplace top-ups and flagged offers that draw credits in non-production | Test and sandbox use now counts against the same pool | UsagePricing | (third-party evidence) |
| 2026-10-12 | Workday | Generally available rate card added Decision Intelligence (8, 24, or 200 credits), Extend agents and tool calls, Sana narration, Data Cloud rows, and a Core Reporting Agent at 8. Frontline upload cut from 15 to 5 credits | More of the platform meters on credits. Rates for generally available skills lock for the order term | UsagePricing | (third-party evidence) |
| 2026-10-13 | IFS | IFS Intelligence sold under a universal enterprise-wide license plus AI Executions, “not a credit model but a fair usage cap” | ERP agent use comes inside the license, with the cap as the limit | IFS, via the ERP edition | (vendor claim) |
| 2026-10-13 | Workday | Autopilot launched with no published price. Usage-based charging reported without rates | The most autonomous tier arrives with no rate to budget against | Workday IR (vendor claim), DiarioBitcoin (third-party evidence), via the AP edition | (vendor claim), (third-party evidence) |
| 2026-10-13 | Causa Prima | $20M seed for agents priced on a cut of negotiated early payments | The first AR agent priced on outcome alone, with no subscription floor disclosed | Fortune | (third-party evidence) |
| 2026-10-16 | Workday | Rising recap reports agents sold on credits with no published dollar price, consumption pricing for Sana Home and Agent Harness, and autopilots to early adopters in early 2027 | Buyers sign for credits before the most expensive agents ship. Analyst advice is to get a dollar price per credit in writing | Beri | (third-party evidence) |
| 2026-10-15 | HubSpot | Customers swap seats for credits mid-term without waiting for renewal. A default account credit limit sits “up to 50% above a customer’s monthly credits” | The seat-to-credit move becomes the buyer’s choice, with a cap on by default | HubSpot, via the AI sales edition | (vendor claim) |
| 2026-10-20 | ZoomInfo | ”Set a Monthly Credit Budget on Each Agent Team” | Spend caps reach the agent team, not only the account | ZoomInfo, via the AI sales edition | (vendor claim) |
| 2026-10-21 | Meta | Muse to keep “a percentage of the commerce” it routes, rate not stated | Agent surfaces add a toll on top of card fees | TechCrunch, via the payments edition | (vendor claim) via (third-party evidence) |
| 2026-10-21 | Stripe | Treasury for agents launched with no published price. Agent spending limits promised “in the coming weeks” | Money movement by agents ships before its price | Stripe, via the payments edition | (vendor claim) |
| 2026-10-23 | Anthropic | Spending caps at organization, group, and member level, pooled group budgets, and an “Efficient mode” limited to Sonnet and Haiku | Model choice becomes a cost control the admin sets | BigGo, via the platforms edition | (third-party evidence) |
| 2026-10-15 | Microsoft | Pay-as-you-go becomes the default for new Copilot Business licenses through CSP resellers from 2026-12-01, with a default limit of 4,000 Copilot Credits per user per month | Agents keep running on the meter instead of stopping at a prepaid pool, up to the default limit | Microsoft Partner Center, October 2026, read through the platforms edition. Corrects the 2026-10-26 aggregator date | (vendor claim) |
| 2026-10-27 | SAP | TechEd. Joule Studio runtime free through December 2026 for selected customers, design-time tools free under fair use. Enterprise AI Gateway due Q1 2027 as an unpriced premium offering | The free period ends as the paid gateway arrives, so the 2027 price event is set | EuropaWire copy of SAP’s release, via the ERP edition | (vendor claim) |
| 2026-10-27 | Microsoft | Copilot Studio billing page revised. Reasoning models bill 10 credits per 1,000 tokens on top of the feature rate, cache reads free. GitHub Copilot harness agents bill separately. Agent flows block new runs when capacity runs out | Agent cost now depends on prompt length and model choice, not message count alone | Microsoft Learn | (vendor claim) |
| 2026-10-27 | Salesforce | Agent Fabric cost controls generally available. Model Wallet sets hard limits in dollars or tokens per application or agent, and blocks requests at the budget. FinOps Dashboard gives one view of AI spend. Cost Optimization Policies claim up to 75% fewer tokens in test scenarios. No price for Agent Fabric | A cost control on top of an unchanged credit price. Enforcement is approximate, so usage can briefly pass the limit | Salesforce, MuleSoft docs | (vendor claim) |
5. Patterns by market
CRM and customer service. This market sets the prices other markets copy. Per-outcome prices run from about $0.50 (HubSpot, third-party report) to $0.99 (Intercom) to $1.50 to $2.00 (Zendesk).
Salesforce charges $2 per conversation or $0.10 per action. HubSpot’s April move from a charge per conversation to one per resolved conversation halved the unit price and moved failure cost to the vendor. Sierra and Decagon publish the model and keep the rate in the contract. Seats stay as the gate: Salesforce $125, Zendesk $50, Microsoft $30.
ERP. ERP vendors sell agents on credits or AI units, and none of the large three publishes a dollar rate. Workday publishes the most detail (credits per action for dozens of skills) and still no price per credit. Oracle sells agentic applications as a separate subscription, and SAP’s AI unit price did not load.
Odoo is the exception, with agentic AI inside its Custom seat. The mid-market splits from the large vendors. IFS and Infor bundle agents under fair use, Unit4 gives AI away to 2027-08-31, and Digits charges firms only for touch-free work. ERP agent cost is therefore a negotiation, not a lookup.
AR automation. AR vendors use the widest spread of models and publish the fewest prices. Tabs includes four agents in a $2,000 platform fee, and Upflow bundles agents by revenue tier. Sidetrade bundles task volume with a compute-price guarantee (GlobeNewswire). Causa Prima takes a success fee, unpriced (Fortune). BlackLine has declared a move to outcome pricing with no rate (transcript).
AP and spend management. Spend platforms include agents in the plan and earn on payments. Ramp and Brex start at $0 per user. BILL includes its W-9 and Invoice Coding agents in $49 to $89 seats and charges on each payment.
Tipalti prices the platform and the payment, not the agent. ERP vendors meter the same AP work instead. NetSuite draws AI Units from a pooled 1,000 per user, and Business Central bills its Payables Agent in Copilot Credits. An agent price in this market competes with free.
AI sales and marketing tools. Prospecting tools price on credits for data plus a meter for work. Clay splits the two (about $0.05 per data credit, about $0.004 per action) and meters token-heavy models by use. Seat-plus-credit plans run from $20 per seat (Unify) to $2,500 a month (Common Room), and ZoomInfo sells credits with no seat fee. Apollo hides its prices from automated reads ($49 to $119 per seat by third-party reads), and Artisan sizes unpublished plans by leads and mailboxes. 11x publishes subscriptions from $2,417 to $5,333 a month by agent and channel. Credits that expire each cycle (Apollo, HubSpot) push the unused value back to the vendor.
Agentic payments. Payment rails charge for the payment, not the agent. Stripe adds $0.15 per Shared Payment Token on top of card fees. Coinbase charges $0.001 per settled x402 payment after 1,000 free, the x402 protocol itself charges nothing, and Visa and Mastercard publish no fee. Meta plans a share of the commerce Muse routes, the first agent surface to state a take. The agent becomes a new way to start a transaction that earns the usual take-rate.
Agent platforms and integration. Integration platforms meter steps. Zapier counts tasks and agent activities, n8n counts whole workflow runs, Lindy and Make sell credits, and Workato quotes. Cloud vendors meter infrastructure (AWS AgentCore by vCPU-hour, Anthropic Managed Agents by session-hour) and pass model tokens through at list. n8n’s per-run unit is the most predictable for buyers, since an agent that takes ten steps costs the same as one that takes two.
Model APIs. Token prices fell again in the newest model generations. Section 7 carries the figures and how vendors pass them through.
6. Who pays when the outcome fails
| Vendor | Failed or unresolved work | Cap or guarantee | Overage rule | Credit expiry or rollover | Source |
|---|---|---|---|---|---|
| Intercom Fin | Charged only on an outcome, but an outcome includes a handoff after a completed Procedure | None stated. Monthly minimum when Fin runs on another helpdesk | Per outcome, uncapped | Not applicable | Intercom (vendor claim) |
| Zendesk AI agents | Free if escalated to a human | Committed rate $1.50 against $2.00 pay-as-you-go | Pay-as-you-go rate above the included allowance | Included resolutions per agent per month | Zendesk (vendor claim) |
| HubSpot Customer Agent | Unresolved chats cost nothing since 2026-04-14 | Default account credit limit up to 50% above monthly credits, from 2026-10-15 | Pay-as-you-go credit packs | Credits reset monthly, no rollover | HubSpot (vendor claim), HubSpot news (vendor claim), eesel (third-party evidence) |
| Sierra | Unresolved and escalated “in most cases” free | Outcome criteria agreed upfront | Contract terms | Contract terms | Sierra (vendor claim) |
| Decagon | Escalations free on per-resolution. Per-conversation charges every conversation | Larger resolution commitments lower the rate | Contract terms | Contract terms | Decagon (vendor claim) |
| Salesforce Agentforce | Every action bills, successful or not | Model Wallet hard limits in dollars or tokens per application or agent, generally available 2026-10-27, with approximate enforcement. Pre-Commit “widely available later this year” | Pay-as-you-go and Pre-Commit overage billed monthly in arrears | Not stated on the page | Salesforce, MuleSoft docs (vendor claim) |
| Microsoft Copilot Studio | Every feature event bills | Per-agent monthly limits set by the admin | Grace to 125% of prepaid capacity, then custom agents disabled. Agent flows block new runs at 100%. Pay-as-you-go removes the block | Pack credits do not roll over. Pre-purchase credits expire at the end of the annual term | Microsoft Learn (vendor claim) |
| Workday Flex Credits | Every action bills | Rates for generally available skills locked for the order term | Alerts at 80, 90, and 100%. Offers are not deactivated, so agents run past the pool. The ERP edition reports a pause-at-limit alert shown at Rising (third-party evidence), which conflicts. Policy and Workday’s blog conflict on how overage is settled | Complimentary credits reset each 1 January, no rollover | UsagePricing, Beri (third-party evidence) |
| ServiceNow assists | Every assist bills | Bundled assist pool per tier | Top-up packs. Reported overage ran 15 to 30% of tier spend | Development instances draw on the production pool | Redress Compliance (third-party evidence) |
| Sidetrade | Bundled task volume | Compute price guaranteed through multi-year contracts | Not public | Not public | GlobeNewswire |
| Zapier | Failed actions are free | Hard stop at 3 times the plan even with pay-per-task on | 1.25 times base (annual) or 2.5 times (monthly) | Allowances reset monthly, even on annual plans | Zapier (vendor claim) |
| n8n | A run that does not finish is not billed (“runs from start to finish”) | Starter and Pro overage capped | Business keeps running, with possible overage charges | Assistant credits do not roll over. No pro-rata refunds on cancellation | n8n (vendor claim) |
| Clay | Empty enrichments cost nothing | None stated | Top-ups at a 30% premium | Actions reset each cycle. Data credit banking conflicts on the page (2 times monthly, or one month, or 15% of annual) | Clay (vendor claim) |
| Apollo | Every credit bills | None stated | Add-on credits charged at once, renewing each cycle | Plan and add-on credits expire each cycle | Apollo (vendor claim) |
| Make | Every credit bills | Auto-purchase of 10,000 credits optional | Bundles of 1,000 or 10,000 at the plan’s fixed price | Extra credits expire after one month (monthly) or at year end (annual) | Make (vendor claim) |
Three findings stand out. First, outcome pricing protects the buyer from failed work but hands the vendor the definition, and Intercom’s definition counts a handoff after a completed Procedure. Second, the large credit vendors split on runaway use. Microsoft shuts agents off, Workday does not, and Salesforce shipped hard caps this month. Third, no vendor in the tracker publishes a refund for a wrong action taken by an agent. Every refund or credit found applies to unresolved work, not to harmful work.
7. The cost floor
Model API prices set the floor under every agent price above. The table holds each provider’s current tiers at standard rates, read 2026-10-30.
| Provider | Model | Input per million tokens | Output per million tokens | Cache hit per million | Notes | Tag |
|---|---|---|---|---|---|---|
| Anthropic | Claude Fable 5.1 | $10 | $50 | $0.25 | Mythos 5.1 at the same rates, limited availability | (vendor claim) |
| Anthropic | Claude Opus 5.5 | $4 | $20 | $0.20 | Opus 5 and 4.x at $5 and $25 | (vendor claim) |
| Anthropic | Claude Sonnet 5.5 | $2 | $10 | $0.10 | Sonnet 5’s $2 and $10 introductory price became standard. The increase to $3 and $15 set for 2026-09-01 will not happen. Cache reads cut from $0.20 on 2026-10-07 (platforms edition) | (vendor claim) |
| Anthropic | Claude Haiku 5.5 | $0.10 | $0.50 | $0.01 | $0.50 and $2.50 for prompts over 100,000 tokens | (vendor claim) |
| OpenAI | gpt-6-astra | $10 | $50 | $1.00 | Short context up to 272,000 tokens | (vendor claim) |
| OpenAI | gpt-6.1-sol | $2 | $10 | $0.10 | (vendor claim) | |
| OpenAI | gpt-6.1-terra | $1 | $5 | $0.05 | (vendor claim) | |
| OpenAI | gpt-6-luna | $0.10 | $0.50 | $0.01 | (vendor claim) | |
| Gemini 3.1 Pro Preview | $2 ($4 above 200,000 tokens) | $12 ($18 above 200,000) | $0.20 | (vendor claim) | ||
| Gemini 3.8 Flash | $0.75 to 2026-12-31, $1.50 from 2027-01-01 | $3.75 to 2026-12-31, $7.50 from 2027-01-01 | $0.075, then $0.15 | Price doubles on 1 January. Gemini 3.6 Flash the same | (vendor claim) | |
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 | $0.025 | (vendor claim) |
Sources: Anthropic, OpenAI (openai.com/api/pricing refused the fetch), Google.
Tool and location charges sit on top of tokens. Web search costs $10 per 1,000 calls at Anthropic and OpenAI, and $14 per 1,000 at Google after 5,000 free a month on Gemini 3. Anthropic’s batch API halves token prices. Anthropic charges 1.1 times for US-only inference, and OpenAI adds 10% for data residency on eligible models.
Vendors pass tokens through in four ways.
| Pass-through | Who | What the buyer sees | Tag |
|---|---|---|---|
| Hidden in a flat unit | Salesforce ($0.10 per action), Intercom ($0.99 per outcome), Zendesk, n8n per run | No token exposure. The vendor keeps the gain as model prices fall and carries the risk of long prompts | (vendor claim) |
| Converted to credits per token | Microsoft (basic 0.1, standard 1.5, premium 10 credits per 1,000 tokens), Clay variable models | At $0.01 per credit, those rates equal $1, $15, and $100 per million tokens, input and output alike (arithmetic). Microsoft’s rates apply to every model it provides, so the spread against list API prices is margin plus model mix, not a measured markup | (vendor claim) |
| Priced by model tier | Zapier AI steps (Standard, BYOM, Advanced, Premium), Lindy credits by task size | The buyer picks the tier. Per-tier rates sit on a page that did not load (zapier.com/pricing/rates returned 404) | (vendor claim) |
| Passed at list or bring your own key | Anthropic and Microsoft marketplaces (consumption units at $0.01 each, no markup stated), AWS AgentCore, n8n self-hosted, Clay with your own keys | The buyer pays the model provider’s list price or its own contract rate | (vendor claim) |
The floor moved down, and two promotions end soon. Opus 5.5 lists at $4 and $20 against $5 and $25 for Opus 5. Gemini 3.8 Flash doubles on 2027-01-01, and OpenAI’s GPT-5.6 Sol promotional price runs “through at least” 2026-11-21. Sidetrade is the only vendor found that guarantees its compute price to the buyer through a multi-year contract (GlobeNewswire).
8. Dated checkpoints
| Date | Event | What to read | Source |
|---|---|---|---|
| Mid November (estimated) | Oracle Fusion 26D in production | The first public AI unit rate for agentic applications | AR Automation Market report |
| On or around 2026-11-21 | Workday Sana Learn narration starts drawing Flex Credits | Whether the 30 credits per minute rate holds | UsagePricing (third-party evidence) |
| 2026-11-21 (at least) | OpenAI GPT-5.6 Sol promotional price ends | The standard price that replaces it | OpenAI (vendor claim) |
| 2026-11-30 | Stripe rejects non-agent API keys for its MCP server | Whether agent credentials get their own price | Stripe docs |
| 2026-12-31 | SAP Joule Studio free runtime ends for selected customers. Unit4 registration for free AI closes | The first SAP runtime rate, and how many Unit4 customers sign | ERP Market report |
| By 2026-12-31 | Salesforce Pre-Commit for Flex Credits (“later this year”) | The discount for committed credits | Salesforce (vendor claim) |
| 2026-12-31 | Microsoft 365 Copilot Business promotion closes | Whether $18 extends or $21 returns | Microsoft (vendor claim) |
| 2027-01-01 | Gemini 3.8 Flash doubles to $1.50 and $7.50 | Whether rivals follow on small-model prices | Google (vendor claim) |
| 2027-01-01 | Stripe’s 0.8% stablecoin rate through Link ends | Whether the promotion extends | Stripe (vendor claim) |
| 2027-01-01 | Workday complimentary credits reset | How much unused value expires | UsagePricing (third-party evidence) |
| Early 2027 | Workday autopilots (Talent Acquisition, Payroll, Accounting) to early adopters | Credit rates for the expensive autonomous tier | Beri (third-party evidence) |
| Q1 2027 | SAP Enterprise AI Gateway | The first price for outside agents reaching SAP data | ERP Market report |
9. Evidence limits
- Corrected 2026-10-30 against the sibling AI sales and marketing edition. The 11x, Artisan, and HubSpot Customer Agent rows first read as quote only, credits, and third-party. The vendors’ own pages show otherwise. The SAP, Dynamics 365 ERP MCP, and AiSDR rows come from the ERP and AI sales editions.
- List prices are not deal prices. Every price above is what a vendor publishes or a third party reports. Enterprise deals carry discounts, commitments, and caps that no public method sees.
- The largest vendors hide the unit price. Workday, Oracle, SAP, and ServiceNow publish units or models and no dollar rate. Those rows rest on third-party reports or carry no price at all.
- Third-party price reports have interests. Many Sierra and Decagon estimates come from rivals (Fin, Quickchat, Featurebase, eesel) that sell competing agents. They are tagged (third-party evidence) and shown as ranges.
- Several pages blocked or hid prices from automated reads. sap.com and openai.com refused fetches. The Apollo, Artisan, and Make pages returned no readable figures, Google’s Gemini Enterprise pricing URL returned 404, and Microsoft’s Copilot Credits licensing page returned an error. A missing price on those rows means “not readable,” not “not published.” Apollo, Make, and Gemini Enterprise now carry third-party prices only.
- Edition rows are second-hand reads. Rows marked with an edition in the Read column come from that market report’s 2026-10-30 reads, cited to its original source where it gave one. They were not re-fetched for this tracker.
- AR rows were not re-read. The Sidetrade, BlackLine, and Causa Prima rows carry 2026-10-18 reads from the AR Automation Market report.
Unverified items.
- SAP AI unit price and what consumes AI units. Not retried, since the ERP edition found sap.com blocked and only snippets carrying €7 and about US$0.40.
- Oracle AI unit rate for agentic applications. Third-party figures now in the row (about $0.01 a unit, a $500,000 SKU snippet), none from Oracle.
- Workday dollar price per Flex Credit. Workday’s page and two Redress guides (updated 2026-08-06 and 2026-09-24) confirm no published rate. Search results carry conflicting advisor ranges ($0.015 to $0.045, $0.15 to $0.22, and an outlier of $50 to $150), none used.
- HubSpot’s $9 per 1,000 annual credit rate (third-party only). The 2026-04-14 start is now confirmed on HubSpot’s own page.
- Which parts of Microsoft’s 2026-10-27 billing page are new in October. Copilot Studio’s What’s new page (updated 2026-10-22) lists no October billing entry, and its August entry introduced usage-based billing for GitHub Copilot harness agents.
- Sierra and Decagon per-unit rates and platform fees. Not retried.
- ServiceNow tier prices and tier assist allocations. The $0.20 rate card rests on Raymond James through Redress, and ServiceNow publishes no price list.
- Gemini Enterprise seat prices. Two third-party sets conflict ($21 and $30 against $25 and $35), and Google’s page was not readable.
- Apollo and Make prices. Third-party reads only, and credit allowances conflict across guides.
- Odoo seat prices. This tracker and the ERP edition read different figures the same day, likely billing period or region.
- Ramp AR add-on price, and Ramp agent pricing beyond “no separate price.”
- Stripe machine-payment fees for MPP and x402, and Stripe Treasury for agents. Not retried.
- Visa and Mastercard agentic fees. Neither discloses one per the payments edition, whose fetch of Mastercard’s newsroom failed.
- Zapier AI step rates per model tier (rates page returned 404, not retried).
- Clay data credit rollover (three conflicting statements on one page) and its FAQ entry prices ($185 and $495 against $167 and $446).
- n8n Pro Assistant credits (9,600 in the plan table, 4,000 in the FAQ).
- Artisan list prices. Not retried.
- Anthropic’s 2026-10-23 spending caps and Google’s 2026-10-08 Gemini agent caps, both through the platforms edition, the first from an aggregator.
How was this study done?
- Method
- Public sources only. Desk study of how vendors price AI agents across six markets. Prices read 2026-10-30 unless a row says otherwise. The study is refreshed monthly, and this edition is the first, the October 2026 edition.
- Sample
- Desk study. 102 public sources: vendor pages, pricing pages and docs, filings and earnings material, analyst and press coverage, and government pages.
- Confidence
- Every claim comes from a public source and carries a link. Vendor claims are marked (vendor claim) and are unverified. Third-party evidence covers analyst reports, surveys, filings, trade press, and government pages. Web search was limited during the research window, and several analyst, review, and vendor pages blocked automated reads, so some claims rest on press releases, wire copies, or search summaries, and each is labeled. The evidence shows what vendors publish and how analysts and the press describe it. It cannot show what vendors win or what agents do in production. The final section lists the items that stay unverified.
Sources
- Fortune
- Salesforce
- Sales pricing (salesforce.com)
- HubSpot
- PricingSaaS
- HubSpot catalog (legal.hubspot.com)
- HubSpot
- Intercom
- Zendesk
- Sierra
- CloudTalk
- Decagon
- Featurebase
- Microsoft
- Pay-as-you-go meters (learn.microsoft.com)
- Microsoft Learn
- Microsoft Learn
- Microsoft
- Microsoft
- Microsoft
- Microsoft Learn
- Microsoft
- Redress tiers (redresscompliance.com)
- Redress assists (redresscompliance.com)
- Workday
- UsagePricing
- Redress (redresscompliance.com)
- Oracle global price list, dated 2026-10-08
- Boston SD
- SAP pricing page
- SAP user-group deck, 2026 (assets.dm.ux.sap.com)
- EuropaWire copy of SAP's release (news.europawire.eu)
- Odoo
- Odoo 20
- IFS (prnewswire.com)
- Unit4 (prnewswire.com)
- ERP Today
- Intuit (investors.intuit.com)
- Software Connect
- GlobeNewswire
- Transcript (investing.com)
- Tabs
- Upflow
- Ramp
- BILL
- product updates (bill.com)
- Zip
- Brex
- Tipalti
- Clay
- Alice (11x.ai)
- Julian (11x.ai)
- Artisan
- AiSDR
- Regie.ai
- GTM.AI
- Common Room
- Unify (unifygtm.com)
- Amplemarket
- Qualified
- Jasper
- Writer
- Klaviyo
- Apollo
- Zeliq
- Docket
- Stripe
- docs
- Stripe
- x402.org
- Stripe
- TechCrunch
- Perplexity
- Coinbase (docs.cdp.coinbase.com)
- Visa (developer.visa.com)
- Zapier
- n8n
- Workato
- Lindy
- Make
- Jetadmin
- UiPath
- AWS
- Anthropic (platform.claude.com)
- Claude pricing
- Ecorpit
- G-Cloud document (assets.applytosupply.digitalmarketplace.service.gov.uk)
- OpenAI (developers.openai.com)
- Google (ai.google.dev)
- TechTarget
- The Register
- Workday IR (investor.workday.com)
- DiarioBitcoin
- Beri
- HubSpot
- ZoomInfo
- BigGo (finance.biggo.com)
- Microsoft Partner Center, October 2026 (learn.microsoft.com)
- Salesforce
- MuleSoft docs
- eesel
- Stripe docs
Published .
Cite this
Couch, C. (2026). Agent Pricing Tracker: October 2026. Ends at Cash. https://chriscouch.ai/research/agent-pricing-tracker-2026/
BibTeX
@misc{couch2026agentpricing,
author = {Couch, Chris},
title = {Agent Pricing Tracker: October 2026},
year = {2026},
howpublished = {Ends at Cash},
url = {https://chriscouch.ai/research/agent-pricing-tracker-2026/}
}