On this page
Model API prices have a dozen trackers. The prices of the agent products built on top of them have none that I could find, and they move more often than people think. Cursor rewrote its Pro plan in June 2025 and apologised 18 days later. Salesforce has priced Agentforce four different ways in under two years. Intercom renamed the unit it bills on without changing the number.
I keep the AI agent pricing index, which answers "what does it cost today". This page answers the next question a buyer asks: when did it change, from what, to what, and in which unit. Every row carries a date and a source, and the whole thing is free to copy.
- 5 verified agent pricing changes in the last 90 days, across 5 vendors. The log holds 38 dated changes from 23 vendors since August 2024, each with a source.
- 4 of 15 billing-unit changes moved toward usage or outcome pricing; 2 moved back toward seats or a flat plan. The other 9 swapped one unit for another in the same family, such as requests for credits.
- 56% of 41 agent products bill on a usage meter: 13 per credit, 6 per run or task, 4 per minute. 4 charge per outcome, 4 per seat, 7 mix a plan price with metered overage and 3 sell a flat subscription.
- $20 a month is the median entry paid price across the 29 products in the census that publish a monthly figure. The lowest is $5.
- 5 increases, 7 decreases. The other 26 restructure what a plan includes or introduce a new unit, so the headline price often stays put while the bill moves.
Updated . Data CC BY 4.0; download the change log CSV and the census CSV.
Cite this page: "AI agent pricing change log and pricing-model census", Gravity, updated , https://gravity.fast/data/ai-agent-pricing-changes/. Data CC BY 4.0: copy the rows into your own work and link back here. Downloads: change log CSV and census CSV.
The change log
One row per change, newest first. "Billing unit" shows the unit before and after where the change moved it, for example requests to credits. "Type" is my label: an increase or decrease means the buyer pays more or less for the same thing; a restructure changes what the plan includes; a new unit is a product's first published meter.
| Date | Vendor and product | What changed | Old → new | Billing unit | Type | Source |
|---|---|---|---|---|---|---|
| Taskade Taskade Enterprise | Enterprise lost its published price: the card read $250 a month with a self-serve Get Enterprise button on 1 September 2026 and now reads Custom with a Contact sales button. Free, Pro, Business and Max are unchanged at $0, $10, $25 and $100 a month on annual billing. The page's own Offer structured data still carried the old $250 a month ($3,000 a year) figure when we checked Write-up. | $250 a month, annual billing, self-serve → Custom, contact sales | flat (unchanged) | Restructure | Wayback Machine / vendor page | |
| Gumloop Gumloop Pro | Gumloop fixed Pro at $37 for 20,000 credits and added an 8% orchestration fee on agent chats, with overage off by default. | Pro slider from $37 (20k credits) to $1,840 (1M credits); unlimited overage by default → Pro $37 per month for 20,000 credits; 8% orchestration fee (16% with BYOK); overage opt-in, capped at 1M credits | credit (unchanged) | Increase | UsagePricing, Gravity pricing index | |
| Lindy | Lindy moved from flat per-account plans to per-user plans with a credit pool shared across the workspace, cutting entry Plus to $29.99 per user. | Plus $49.99 per month per account → Plus $29.99 per user per month (3,000 pooled credits); Pro $99.99; Max $199.99 | flat to seat | Restructure | UsagePricing, Lindy Docs, Gravity pricing index | |
| Make Make plans | Make collapsed its Core, Pro and Teams tiers into one paid plan with a lower 5,000-credit entry step. | Core $9, Pro $16, Teams $29 per month (annual) at a 10,000-credit entry step → One paid Make plan from $9 per month (annual) at 5,000 credits | credit (unchanged) | Restructure | UsagePricing, Gravity pricing index | |
| Cursor Cursor Start (India) | Cursor launched an India-only plan priced in rupees and payable by UPI. | Cheapest paid individual plan: Pro at $20 per month → Cursor Start at INR 649 per month, tax inclusive | flat (unchanged) | Decrease | Cursor | |
| Salesforce Agentforce Help Agent | Salesforce announced a packaged service agent billed only when it resolves an issue on its own, available from July 2026. | $2 per conversation or $0.10 per action → $2 per resolution | conversation to resolution | New unit | Constellation Research, Salesforce Ben | |
| Cursor Cursor Teams | Cursor split Teams usage into first-party and third-party pools and added a Premium seat with 5x the Standard seat's usage at 3x the cost, for renewals from July 1, 2026. | Standard seat $40 per month → Standard seat $40 per month ($32 annual) with more included usage; Premium seat $120 per month ($96 annual) | seat (unchanged) | Restructure | Cursor | |
| GitHub GitHub Copilot (all plans) | GitHub replaced premium request units with GitHub AI Credits, moving every Copilot plan to usage-based billing at unchanged seat prices (announced 2026-04-27). | Premium request allowances; $0.04 per extra premium request → Monthly AI Credits included with each plan (1 credit = $0.01); Pro $10, Pro+ $39, Business $19/user, Enterprise $39/user | request to credit | Restructure | GitHub Blog, GitHub | |
| Lovable | Lovable merged its separate dollar-denominated Cloud and AI balances into the single credit balance used for building. | Build credits plus separate dollar balances for Cloud and AI → One credit balance for build, chat, Cloud, AI gateway and connectors | credit (unchanged) | Restructure | Lovable Docs, UsagePricing | |
| Zapier Zapier (Zaps, AI steps, Agents) | Zapier moved AI steps, code, MCP and SDK usage into the same task pool, at variable task rates per action, with plan prices unchanged. | One task per external-app action; AI steps, code and SDK metered separately → One shared task pool with variable per-action task rates | task (unchanged) | Restructure | Zapier, UsagePricing, Gravity pricing index | |
| Zendesk Zendesk AI agents | Zendesk replaced a single automated-resolution charge with resolution tiers priced by the value the AI agent provided. | One automated resolution unit per resolved request → Automated resolutions grouped into value tiers; assisted escalations do not count against the allowance | resolution (unchanged) | Restructure | Zendesk Help | |
| HubSpot Breeze Customer Agent | HubSpot moved Customer Agent from a per-conversation charge to a per-resolved-conversation charge, paid in HubSpot Credits. | $1.00 per conversation → $0.50 per resolved conversation (50 HubSpot Credits) | conversation to resolution | Decrease | HubSpot, MarTech | |
| HubSpot Breeze Prospecting Agent | HubSpot replaced a recurring monthly charge per enrolled contact with a charge per lead the agent recommends for outreach. | Recurring monthly charge for every contact enrolled → $1.00 per lead recommended for outreach (100 HubSpot Credits) | task (unchanged) | Restructure | MarTech | |
| Windsurf (Cognition) Windsurf Pro, Teams, Max | Windsurf replaced monthly credits with daily and weekly quotas, raised Pro from $15 to $20 and added a $200 Max tier. | Pro $15 per month with 500 monthly credits → Pro $20 per month with daily and weekly quotas; Teams $40 per seat; Max $200; extra usage at API pricing | credit to flat | Increase | Cognition (Devin blog), CloudZero | |
| Intercom Fin AI Agent | Intercom changed Fin's billing unit from resolutions to outcomes, which also bill configured Procedure handoffs, at the same $0.99. | $0.99 per resolution → $0.99 per outcome (resolutions, Procedure handoffs, disqualifications, self-serve routings); $9.99 per sales qualification | resolution (unchanged) | Restructure | Intercom, Fin (Intercom) | |
| Gumloop Gumloop Pro | Gumloop merged its Solo and Team plans into one Pro plan with shared credits and no seat limits. | Separate Solo and Team plans with seat limits → One Pro plan, shared credits, no seat limits, more credits at the same price | seat to credit | Decrease | Gumloop | |
| Replit Replit Core and Pro | Replit cut Core from $25 to $20 a month, added a $100 Pro plan with tiered credits, and retired the Teams plan. | Core $25 per month; Teams plan → Core $20 per month (includes $20 of credits); Pro from $100 per month | flat (unchanged) | Decrease | Replit | |
| Relevance AI | Relevance AI split its single credit into Actions (agent work) and Vendor Credits (model costs passed through without markup). | One credit covering both agent work and model costs → Actions that reset each cycle plus Vendor Credits at cost that roll over | credit to action | Restructure | Relevance AI | |
| Microsoft Copilot Studio | Microsoft renamed Copilot Studio messages to Copilot Credits, metered at feature-based rates, with pack price and pay-as-you-go rate unchanged. | $200 per 25,000-message pack per month; $0.01 per message pay-as-you-go → $200 per 25,000-Copilot-Credit pack per month; $0.01 per credit pay-as-you-go | message to credit | Restructure | Microsoft Learn, A Guide to Cloud, Microsoft | |
| Anthropic Claude Pro and Max (Claude Code) | Anthropic added weekly rate limits on top of the five-hour limits for Pro and Max subscribers (announced 2025-07-28). | Five-hour rolling limits only → Five-hour limits plus an overall weekly limit and an Opus weekly limit; same $20/$100/$200 prices | flat (unchanged) | Increase | TechCrunch | |
| Make Make (automation platform with AI agents) | Make switched its billing unit from operations to credits at 1:1, with AI modules priced by actual usage. | Billed in operations (one per module run) → Billed in credits; most module runs 1 credit, AI features at fixed or token-based rates | action to credit | Restructure | Make Help Center, Make Community | |
| Cursor Cursor Teams and Auto | Cursor moved Teams agent usage to variable request costs and ended unlimited Auto for individuals, counting Auto against included usage from renewals after September 15, 2025. | Teams: fixed request costs; Auto unlimited for individuals → Teams: variable request costs; Auto draws on included usage at token rates | request to credit | Increase | Cursor | |
| n8n n8n Cloud and Enterprise | n8n removed active-workflow limits from every plan and moved Enterprise pricing from workflow count to executions, adding a self-hosted Business plan. | Plans capped by active workflows; Enterprise priced by workflow count → Unlimited workflows, steps and users on every plan; billed by executions | execution (unchanged) | Restructure | n8n | |
| Activepieces Activepieces plans | the page's plan set changed to $0, $25 and $150 a month, per our July 2025 snapshot; by September 2026 Plus is $16 a month yearly with a per-credit overage | n/a → $0, $25 and $150 a month | n/a | Restructure | Wayback Machine / vendor page | |
| StackBlitz Bolt.new | Bolt let unused paid-plan tokens roll over for one extra month. | Unused monthly tokens expired → Unused tokens on paid plans valid for two months | credit (unchanged) | Decrease | Taskade, Bolt Support | |
| GitHub GitHub Copilot (Pro, Pro+, Business, Enterprise) | GitHub began enforcing monthly premium-request allowances, with extra premium requests billed per request. | Premium request allowances announced but not enforced → Pro 300, Pro+ 1,500, Business 300, Enterprise 1,000 premium requests a month; extra at $0.04 each | seat to request | Increase | GitHub Changelog, The Register | |
| Replit Replit Agent | Replit switched Agent from a flat per-checkpoint fee to effort-based pricing, one checkpoint per request priced by the work done (new users first, Core and Teams from July 1). | $0.25 per checkpoint → One checkpoint per request, priced by effort (below or above $0.25) | task (unchanged) | Restructure | Replit | |
| Salesforce Agentforce add-ons and Agentforce 1 Editions | Salesforce made per-user Agentforce add-ons generally available, giving licensed employees unmetered agent use. | Consumption only (per conversation or per action) → Add-ons from $125 per user per month (unmetered for licensed users); Agentforce 1 Editions from $550 per user per month | action to seat | New unit | Salesforce, The Register | |
| Cursor Cursor Pro | Cursor moved Pro from a monthly request limit to a dollar pool of model usage at API prices; after backlash it apologised on 2025-07-04 and refunded surprise charges. | 500 requests a month for $20 → $20 of model inference at API prices a month for $20 | request to credit | Restructure | Cursor, Cursor | |
| Cursor Cursor Ultra | Cursor added a $200 Ultra plan with 20 times Pro usage. | n/a → $200 per month, 20x Pro usage | credit (unchanged) | Restructure | Cursor | |
| Salesforce Agentforce (Flex Credits) | Salesforce added Flex Credits, billing Agentforce per action instead of per conversation. | $2 per conversation → $0.10 per action (20 Flex Credits; credits sold at $500 per 100,000) | conversation to action | Restructure | Salesforce, The Register | |
| Anthropic Claude Max (includes Claude Code) | Anthropic added Max plans above the $20 Pro plan. | Top individual plan: Pro at $20 per month → Max at $100 per month (5x Pro usage) or $200 per month (20x Pro usage) | flat (unchanged) | Restructure | CNBC | |
| GitHub GitHub Copilot Pro+ | GitHub added a higher individual tier, Copilot Pro+, with 1,500 premium requests a month. | n/a → $39 per month with 1,500 premium requests | request (unchanged) | Restructure | GitHub Changelog, GitHub Blog | |
| Cognition Devin | Devin 2.0 replaced the $500-a-month team entry price with a $20 Core plan and pay-as-you-go Agent Compute Units. | $500 per month (ACUs at $2) → Core from $20 (about 9 ACUs), then $2.25 per ACU | ACU (unchanged) | Decrease | TechCrunch, Cognition | |
| Windsurf (Codeium) Windsurf Pro and Teams | Windsurf removed flow action credits so only prompts are charged, and cut Teams pricing. | Prompt credits plus flow action credits for background tool calls; Teams $35 per user per month → Prompt credits only (Pro stays $15); Teams $30 per user per month | action to credit | Decrease | TechCrunch | |
| Manus | Manus launched paid credit plans after its invite-only debut. | n/a → Starter $39 per month (3,900 credits); Pro $199 per month (19,900 credits) | credit (unchanged) | New unit | WinBuzzer | |
| Salesforce Agentforce | Agentforce reached general availability priced per conversation. | n/a → $2 per conversation, with volume discounts | conversation (unchanged) | New unit | MarTech, Salesforce Ben | |
| Zendesk Zendesk AI agents | Zendesk introduced outcome-based pricing, charging per automated resolution by an AI agent. | n/a → Pay per automated resolution ($1.50 committed, $2.00 pay-as-you-go, per eesel) | resolution (unchanged) | New unit | CX Today, Computer Weekly, eesel AI |
38 changes, newest first. Dates are the effective date where the vendor gave one, otherwise the announcement date; a month alone means the source gave no day.
How agent products charge: the census
Each product is classified by the unit its bill scales with on the entry paid plan. A plan that bundles a usage allowance and then meters overage is "hybrid"; a credit is a vendor currency whose burn per action varies, while "per run or task" counts a fixed unit of work. The 16 workflow and agent platforms come from the pricing index read on 15 September 2026, the voice platforms from its 4 October 2026 read, and the rest from each vendor's page on the date shown. Products with a price history page link to it.
| Product | Pricing model | What the bill scales with | Entry price | Read on |
|---|---|---|---|---|
| Amazon Web Services Amazon Q Developer | per seat | users | Pro $19 per user per month | 4 October 2026 |
| Microsoft Microsoft 365 Copilot Business | per seat | users | $18 per user per month paid yearly (promo, from $21); $25.20 monthly | 4 October 2026 |
| Vellum | per seat | seats plus usage | $30 a month (first paid tier on the page) | 15 September 2026 |
| Voiceflow | per seat | per agent and per seat, per the product docs | Not published on the pricing page (contact sales) | 15 September 2026 |
| Activepieces | per run or task | sold as credits, one per run | $16 a month (Plus, yearly billing; $20 on monthly billing) | 15 September 2026 |
| Botpress | per run or task | AI spend per message on top of a monthly plan | $25 a month (first paid tier on the page) | 15 September 2026 |
| Decagon Decagon AI agents | per run or task | conversations by default, resolutions optional (undisclosed rates) | Undisclosed; per conversation by default, per resolution optional | 25 September 2026 |
| n8n | per run or task | executions, one per full workflow run | €20 a month (Starter, annual billing) | 15 September 2026 |
| Stack AI | per run or task | runs a month, plus seats | Paid tiers not shown with a figure on the page | 15 September 2026 |
| Zapier | per run or task | tasks, one per step | $19.99 a month (Professional, annual billing) | 15 September 2026 |
| Bardeen | per credit | credits per action, per the product docs | Not published on the pricing page as read | 15 September 2026 |
| Dify | per credit | message credits | $590 a year on the page's first paid tier ($159 a month on the next) | 15 September 2026 |
| Gumloop | per credit | credits per node run | $37 a month (Pro) | 15 September 2026 |
| Lindy | per credit | credit bucket on a per-seat plan | $29.99 per seat a month (Plus) | 15 September 2026 |
| Lovable | per credit | credits per request | Pro from $25 per month; top-ups $15 per 50 credits | 9 June 2026 |
| Make | per credit | credits, one per module action | $9 a month for 5,000 credits (Make Plan, annual billing; $10.59 monthly) | 15 September 2026 |
| Manus | per credit | credits per task | Basic $19 per month for 1,900 credits | 14 September 2026 |
| Microsoft Copilot Studio | per credit | Copilot Credits, by feature | $200 per pack of 25,000 Copilot Credits per month; $0.01 per credit pay-as-you-go | 4 October 2026 |
| Pipedream | per credit | credits per execution, by compute time | $29 a month (first paid tier, annual billing) | 15 September 2026 |
| Relevance AI | per credit | credits per action, per the product docs | Not published on the pricing page (sign-in or contact) | 15 September 2026 |
| Salesforce Agentforce (Flex Credits) | per credit | actions, sold as Flex Credits | $500 per 100,000 Flex Credits ($0.10 per action); also $2 per conversation or $125 per user per month add-on | 4 October 2026 |
| StackBlitz Bolt.new | per credit | tokens | Pro $25 per month for 10M tokens | 10 April 2026 |
| Vercel v0 | per credit | credits on a per-user plan | Plus $30 per user per month with $30 of included credits | 4 October 2026 |
| Bland Bland AI voice agents | per minute | talk minutes, one bundled rate | $0.14 a minute, $0 platform fee (Start) | 4 October 2026 |
| ElevenLabs ElevenLabs Agents | per minute | monthly bucket of call minutes | $6 a month (Starter, 75 minutes) | 4 October 2026 |
| Retell AI Retell AI voice agents | per minute | talk minutes, itemised by component | $0.07 to $0.31 a minute, pay as you go | 4 October 2026 |
| Vapi Vapi voice agents | per minute | each component billed per minute at cost | Components at cost, roughly $0.040 to $0.087 a minute on inbound US calls | 4 October 2026 |
| HubSpot Breeze Customer Agent | per outcome | each resolved conversation, paid in HubSpot Credits | $0.50 per resolved conversation (50 HubSpot Credits) | 14 April 2026 |
| Intercom Fin AI Agent | per outcome | each outcome: resolution, handoff or qualification | $0.99 per outcome; 50-outcome monthly minimum when standalone | 4 October 2026 |
| Salesforce Agentforce Help Agent | per outcome | each resolution | $2 per resolution | 4 October 2026 |
| Sierra Sierra agents | per outcome | successful outcomes (contract terms undisclosed) | Undisclosed; charged per successful outcome | 3 August 2026 |
| Augment Code | hybrid | plan price with included usage, then top-ups | Standard $20 per month flat for up to 50 seats with $20 of usage; pay-as-you-go top-ups | 4 October 2026 |
| Cognition Windsurf / Devin Desktop | hybrid | plan price with quotas, then usage at API rates | Pro $20 per month with quotas; extra usage at API pricing | 4 October 2026 |
| Cursor | hybrid | plan price with included model usage, then usage at API rates | Pro $20 per month including $20 of model usage at API rates; INR 649 Start plan in India | 4 October 2026 |
| GitHub GitHub Copilot | hybrid | plan price with included AI credits, then usage | Pro $10 per month with 1,500 AI credits (1 credit = $0.01); Pro+ $39; Max $100 | 4 October 2026 |
| OpenAI Codex (ChatGPT plans) | hybrid | ChatGPT plan with included usage, then credits | Go $8 per month; Plus $20 with extra usage via ChatGPT credits | 4 October 2026 |
| Replit Replit Agent | hybrid | plan price with included credits, then effort-priced checkpoints | Core $20 per month ($18 annual) including $20 of credits | 4 October 2026 |
| Zendesk Zendesk AI agents | hybrid | agent seats with a resolution allowance, then per resolution | Suite Team $55 per agent per month (annual) with 5 automated resolutions per agent; extra resolutions $1.50 committed or $2.00 pay-as-you-go | 17 September 2026 |
| Anthropic Claude (Claude Code) | flat subscription | plan with usage limits | Pro $20 per month ($17 annual); Max $100 / $200 | 4 October 2026 |
| Gravity | flat subscription | monthly plan with usage included; buy more usage if it runs out | Basic free (one agent); Pro $5 a month during the alpha; Max $20 a month | 3 October 2026 |
| Taskade | flat subscription | tiered plans with optional top-up packs | $10 a month (first paid tier, annual billing) | 15 September 2026 |
| Synthflow Synthflow voice agents | not published | scoped around call volume, concurrency and telephony | Not published; enterprise contracts from $30,000 a year | 4 October 2026 |
42 products: per seat 4 · per run or task 6 · per credit 13 · per minute 4 · per outcome 4 · hybrid 7 · flat subscription 3 · not published 1. Classified by the unit the bill scales with on the entry paid plan. Euro prices stay in euros.
The census includes Gravity because it is an agent product with a public price: Basic is free with one agent, Pro is $5 a month during the private alpha and Max is $20 a month. It is counted as a flat subscription like any other row.
What the changes have in common
Three patterns hold across the log. Each one is a reading of dated rows above, so check the rows before you quote the reading.
Outcome pricing starts in customer support, and stays there for now
Zendesk began charging per automated resolution in August 2024. Intercom moved Fin from $0.99 per resolution to $0.99 per outcome on 12 March 2026, which also bills configured handoffs. HubSpot moved its Customer Agent from $1.00 per conversation to $0.50 per resolved conversation on 14 April 2026, and Salesforce announced a Help Agent at $2 per resolution on 26 June 2026.
Every outcome-priced product in the census answers support tickets. A resolved ticket is easy to count and both sides agree on what it is worth. Most other agent work has no such unit yet, which is why credits dominate everywhere else.
Coding agents swapped request counts for dollar pools
Cursor replaced 500 requests a month with $20 of usage at API prices on 16 June 2025. GitHub began billing premium requests at $0.04 each beyond the allowance on 18 June 2025, then replaced them with AI credits worth $0.01 each on 1 June 2026, keeping seat prices.
Windsurf is the exception that went the other way. On 19 March 2026 it dropped credits for daily and weekly quotas and raised Pro from $15 to $20 a month. A quota is easier to understand and harder to budget past, so expect the two approaches to keep trading places.
Price rises arrive as a new unit more often than a new number
As of 4 October 2026, only 5 of the 38 changes are plain list-price increases. Many of the rest change what a unit buys: Make replaced operations with credits on 27 August 2025, Relevance AI split one credit into actions and vendor credits on 8 September 2025, and Gumloop added an 8% orchestration fee in August 2026. A buyer comparing headline prices across those dates sees no change at all.
The large cuts are real too. Devin's entry price fell from $500 a month to a $20 Core plan plus $2.25 per compute unit on 3 April 2025, and Replit cut Core from $25 to $20 on 24 February 2026. If you want the monthly-bill view of all this, AI agent pricing explained walks through it, and the cheapest AI agent platforms ranks the floors.
What the pricing pages show between announcements
Not every change gets a blog post. For the 15 platforms the weekly watch snapshots, this table counts how often the set of prices printed on the pricing page moved, from monthly Wayback Machine captures and live reads. It is a signal, not a log: a page that switches its default from monthly to annual figures counts as a movement.
| Tool | Snapshots | Window | Price-set movements | In the last 12 months | Last movement |
|---|---|---|---|---|---|
| Taskade | 32 | 2024-01-03 to 2026-09-11 | 14 | 9 | September 2026 |
| Botpress | 19 | 2024-03-02 to 2026-09-15 | 12 | 6 | August 2026 |
| Activepieces | 28 | 2024-01-09 to 2026-09-11 | 8 | 3 | September 2026 |
| Relevance AI | 27 | 2024-03-13 to 2026-05-13 | 7 | 4 | May 2026 |
| Voiceflow | 18 | 2024-01-08 to 2026-01-06 | 7 | 0 | July 2025 |
| Gumloop | 27 | 2024-06-12 to 2026-09-15 | 5 | 1 | April 2026 |
| Lindy | 20 | 2024-04-13 to 2026-09-15 | 5 | 3 | September 2026 |
| Vellum | 15 | 2024-05-26 to 2026-09-15 | 5 | 5 | September 2026 |
| Bardeen | 18 | 2024-01-09 to 2025-11-12 | 4 | 0 | March 2025 |
| n8n | 23 | 2024-01-08 to 2026-09-15 | 4 | 1 | September 2026 |
| Dify | 30 | 2024-01-13 to 2026-09-15 | 3 | 1 | July 2026 |
| Pipedream | 25 | 2024-01-12 to 2026-09-15 | 3 | 0 | May 2025 |
| Stack AI | 18 | 2024-01-14 to 2026-09-11 | 2 | 0 | May 2025 |
| Zapier | 19 | 2025-05-04 to 2026-09-15 | 2 | 1 | July 2026 |
| Make | 2 | 2024-01-28 to 2025-02-08 | 0 | 0 | none seen |
34 price-set movements across these tools in the last 12 months, and 14 raw page diffs flagged by the weekly watch in the last 90 days. A movement is any change in the set of prices printed on the page, so it includes display changes (annual against monthly figures); only confirmed moves enter the change log above.
Update log
What moved on this page, by date. Corrections are logged here rather than edited in silently.
- : Tracker launched with 38 verified changes from 23 vendors and a census of 42 agent products.
Methodology
What counts. A change to the published price, plan structure or billing unit of an AI agent product: a product that carries out tasks on its own, including workflow platforms with agent features, coding agents, customer-service agents and voice agents. Model API token prices are out of scope; they have their own trackers. A product launch counts when it is the product's first published meter.
What does not count. A pricing page that only changes how it displays a price (annual against monthly figures, a new currency toggle), temporary promotions, and enterprise contract terms nobody publishes.
Sources. For the 16 platforms in the pricing index, a scheduled job snapshots the vendor pricing pages every Tuesday and compares them with monthly Wayback Machine captures back to January 2024; a raw page diff only enters the log after the change is confirmed on the live page. For other vendors, each row cites the vendor's announcement, help centre or pricing page, or credible press. Rows dated from a third-party pricing-page capture carry month precision, because the capture date is not the announcement date.
Dates. The effective date where the vendor gave one, otherwise the announcement date. A month without a day means the source gave no day.
Units and families. Seats and flat plans count as fixed pricing; requests, runs, tasks, executions, messages, actions, credits, compute units, minutes and conversations count as usage; resolutions and outcomes count as outcome pricing. A change "moves toward usage or outcome" when its unit after the change sits in a later family than before.
Known gaps. The census is 42 products, not the whole market, and leans toward products with public pricing. Enterprise agents priced by contract (Sierra, Decagon) appear with undisclosed rates. A few rows rest on secondary sources and are marked as such in the CSV.
Cadence. Rebuilt weekly from stored data. The sibling SaaS AI price increase tracker logs the other side of the same story: existing software getting more expensive because AI was added.
FAQ
How are AI agents priced in 2026?
Mostly on a usage meter. Of the 41 agent products in this census whose model is published, 23 bill on usage (13 per credit, 6 per run or task, 4 per minute), 7 mix a plan price with metered overage, 4 charge per outcome, 4 per seat and 3 sell a flat subscription. Counts as of 4 October 2026.
Which AI agents charge per resolution or per outcome?
Intercom Fin at $0.99 per outcome, Salesforce's Agentforce Help Agent at $2 per resolution (announced 26 June 2026), HubSpot's Customer Agent at $0.50 per resolved conversation (from 14 April 2026) and Sierra, whose rates are undisclosed. Zendesk bills automated resolutions on top of agent seats. Outcome pricing is almost entirely a customer-service phenomenon so far.
Are AI agent prices going up or down?
Both, and the direction is often hidden in the unit. Of the 38 changes logged here through 4 October 2026, 5 are plain increases and 7 are decreases; the rest restructure what a plan includes or introduce a new unit. Devin's entry price fell from $500 to $20 a month in April 2025, while Windsurf raised Pro from $15 to $20 in March 2026.
What is the median price of an AI agent?
$20 a month, the median entry paid price across the 29 products in this census that publish a monthly figure, as of 4 October 2026. Per-minute voice platforms and contract-only products are excluded from the median because they publish no monthly figure.
How often is this tracker updated?
Weekly. A scheduled job snapshots 16 vendor pricing pages every Tuesday and every confirmed change enters the log with its date and source. Changes at vendors outside that set are added when they are announced and verified. Each update is listed in the update log.
Can I reuse this data?
Yes. The change log and census are published under CC BY 4.0 as CSV files on this page. Reuse them with a link back to this page. If a change is missing, send it through contact with a dated source.
Sources
- Salesforce takes a run at outcome-based Help Agent pricing, Constellation Research, 26 June 2026.
- Salesforce pricing update, Salesforce, 17 June 2025.
- From resolutions to outcomes, Intercom, 12 March 2026.
- Customer Agent and Prospecting Agent: now you pay when the task is complete, HubSpot, 14 April 2026.
- June 2025 pricing update and refunds, Cursor, 4 July 2025.
- GitHub Copilot is moving to usage-based billing, GitHub, 27 April 2026.
- Introducing our new Windsurf pricing plans, Cognition, March 2026.
- Devin gets a new pay-as-you-go plan, TechCrunch, 3 April 2025.
- Credits as the new billing unit in Make, Make Help Center, August 2025.
- Splitting the bill, Relevance AI, 8 September 2025.
- Every other row links its own source in the change log and census tables above; full URLs are in the CSV files.
- My weekly pricing-page snapshots and monthly Wayback Machine captures, January 2024 to October 2026, summarised in the AI agent pricing index.
More on what agents cost: AI agent pricing trends in 2026, AI agent pricing statistics, how much an AI employee costs and the Taskade Enterprise change write-up. Our own plans are on the pricing page.