AI companies in 2026 charge in six main ways: per seat (Microsoft 365 Copilot, $30 a user), seat plus a usage allowance (Claude Enterprise, $20 plus usage at API rates), tokens (GPT-6.1 Sol, $2 and $10 per million), credits (Lovable, from $25 for 100), per unit of work (Deepgram, $0.0043 a minute) and per outcome (Fin, $0.99 a resolution). Many also bundle AI into existing plans.
Key takeaways
- Pure per-seat AI pricing survives mainly where the vendor owns the workflow or sells against expensive labor.
- The most common pattern is a plan or seat with an included allowance, then credits or overage.
- Credits dominate creative and app-building tools because task costs vary so widely.
- Outcome pricing works best in customer support, where a resolution can be counted and audited.
- Infrastructure companies publish per-unit rates, and the best ones add a monthly minimum.
In the book, I called Amazon the ultimate moat. Not because of any one thing it sells, but because it combines every payment model I covered: e-commerce, subscription (Prime), advertising, transaction fees from other sellers, and cloud services through AWS.
"You don't have to stick to one model," I wrote. I even said it twice.
AI companies have taken that advice further than I ever imagined. Salesforce now sells its AI agents per action, per conversation, per resolution, per user, by edition and on fixed-price enterprise agreements, all at the same time. ICONIQ found the average AI company blends 1.7 pricing models.
So rather than theory, here's the menu. Forty real companies, sorted into seven models, with what each charges and the one lesson I'd take from it.
Prices checked 27 October 2026; they change often, so see the AI Pricing Index for the live tracker. Figures marked "reported" are third-party estimates, not published prices. Each company name links to its full pricing teardown.
Model #1: Per seat
The classic SaaS price, defended by the companies that own the workflow or sell against expensive human time.
| # | Company | What they charge | Value metric | The lesson |
|---|---|---|---|---|
| 1 | Microsoft 365 Copilot | $30 per user (Enterprise); Business $21, promo $18 to 31 Dec 2026 | Seat | Microsoft bundles AI into its $99 E7 suite rather than metering. Bundling works when you own the workflow. |
| 2 | Jasper | Pro $69 monthly or $59 annual per seat | Seat with feature limits | A $59 seat competes with $20 assistants, so it needs brand voice and knowledge the general tools lack. |
| 3 | Writer | Starter $39 monthly or $29 annual per user | Seat; free Lite seats on Enterprise | Charge the people who create value, and let viewers in free so you spread inside the company. |
| 4 | Slack | Business+ $15 annual (up from $12.50 when AI was added) | Active user | Adding AI is a defensible moment to raise list price, if the value is visible on day one. |
| 5 | Hebbia | Reported $3,000 to $15,000 per seat per year | Seat | Finance research AI is priced like a data terminal. Compare against the incumbent tool, not ChatGPT. |
Model #2: AI bundled into the existing plan
No separate AI price at all. The AI is the reason to stay, or the reason for this year's price rise.
| # | Company | What they charge | Value metric | The lesson |
|---|---|---|---|---|
| 6 | Shopify | Basic $39 monthly; Sidekick AI included on all plans | Subscription plus a share of payments | When you earn a share of customer revenue, give away AI that grows that revenue. |
| 7 | Xero | Comprehensive A$107 a month (from A$100 on 1 Jul 2026) | Organization, tiered by payroll size | AI features bundled in, with small annual increases. Steady beats big jumps for small-business software. |
| 8 | MYOB | AI BAS (beta) add-on A$25 a month on Lite and Pro | Add-on for one compliance job | Bundle general AI, but charge separately for AI that does one specific, valuable job. |
| 9 | Canva | Pro US$144 a year with up to 200 Premium or 20 Ultra AI uses a month | Person, with AI-use caps by tool class | According to PricingSaaS, Canva raised Pro to $180, reverted to $144, then cut free AI by 90%. Test changes on new customers first. |
| 10 | Duolingo | Max about $167.99 a year (reported) | Account, tiered by AI features | Put AI in a tier above your existing paid tier instead of raising the base price. |
Model #3: Seat or plan, plus an allowance
The 2026 default. A price customers recognize, an included pool of usage, and a way to pay for more. If you only copy one model on this page, copy this one.
| # | Company | What they charge | Value metric | The lesson |
|---|---|---|---|---|
| 11 | ChatGPT Business | Standard $20 to $25, Premium $100 to $125 per seat, plus credits | Seat with usage allowance | Mixing standard and premium seats lets heavy users pay more without a separate product. |
| 12 | Claude Team and Enterprise | Enterprise: $20 per seat plus usage at API rates, 20-seat minimum | Seat plus metered usage | Splitting the seat into an access fee and a usage line keeps entry cheap while heavy users pay their cost. |
| 13 | GitHub Copilot | Pro $10 with $15 of AI credits; Pro+ $39; Max $100 | Seat with dollar AI credits | Prices held, allowances changed. Developers said "you will get less, but pay the same price." |
| 14 | Cursor | $8, $30, $100 or $200 via the X Subscription | Weekly usage pool shared with Grok | According to x.ai, Cursor's plans folded into SpaceXAI's bundle on 12 Oct 2026. Bundles work only if buyers value every part. |
| 15 | Replit | Core $20, Pro $100, with included credits | Effort per agent request (time and compute) | Effort pricing tracked costs but surprised customers. Show an estimate before the work runs. |
| 16 | Notion | Business $20 annual per seat with 1,500 Standard and 500 Select AI credits | Seat with two credit classes | Notion retired a $10 AI add-on and used AI to move buyers up a tier instead. |
| 17 | Atlassian | 25, 70 or 150 Rovo credits per user a month, by plan | Seat with pooled AI credits | Pooling credits across a whole site lets light users subsidize heavy ones, so nobody hits a wall. |
| 18 | Figma | Full seat $16 with 3,000 AI credits; 5,000 extra for $120 to $150 | Seat with AI credits | Enforcing limits after a free period sparked a backlash. Warn early and show usage first. |
Model #4: Tokens
Pay for what the model reads and writes. Perfect for developers, confusing for almost everyone else. The LLM API pricing comparison has every model side by side.
| # | Company | What they charge | Value metric | The lesson |
|---|---|---|---|---|
| 19 | OpenAI API | GPT-6.1 Sol $2 in / $10 out; GPT-6 Luna $0.10 / $0.50 per million | Tokens | A 100x price ladder plus 50% off batch jobs. Route each task to the cheapest model that passes. |
| 20 | Anthropic API | Claude Opus 5.5 $4 / $20 per million (down from $15 / $75 for Opus 4) | Tokens, with cache rates | Capability keeps getting cheaper. Price on value and treat falling costs as margin you earn back. |
| 21 | Google Gemini API | Gemini 4 Argon $2 / $10, rising to $4 / $20 on 1 Dec 2026 | Tokens, with launch pricing | Introductory API pricing is a land-grab. Cost your product at the post-increase rate. |
| 22 | Bolt.new | Pro from $25 with 10 million tokens a month | Token allowance | Raw tokens are honest but hard for non-developers to translate. Add an "average project uses X" guide. |
Model #5: Credits
A private currency with a rate card. The dominant model for creative tools and app builders, because one product's tasks can vary 100-fold in cost. The credit-based pricing guide shows how to design one.
| # | Company | What they charge | Value metric | The lesson |
|---|---|---|---|---|
| 23 | Lovable | Pro from $25 for 100 credits | Credits per message | It prints the credit cost of example prompts on the pricing page. That's the gold standard for transparency. |
| 24 | ElevenLabs | $6 to $990 a month; Pro $99 for 600,000 credits | Credits (about one per character) | Rollover for two months, and voice agents at $0.08 a minute on every tier. Sell capacity, not cheaper units. |
| 25 | Runway | Standard $15 for 625 credits; Max $95 for 9,500 | 12 credits per second of video | Rollover only on the top tier makes a clean upgrade reason for heavy users. |
| 26 | Adobe Firefly | $9.99 for 2,000 credits up to $199.99 for 50,000 | Credits for premium features only | Standard generations stay unlimited, so the core feels free while costly models are metered. |
| 27 | Perplexity | Pro $20; Max $200 with 10,000 credits a month | Subscription plus dollar-pegged credits (100 = $1) | Keep chat flat and meter only the expensive agent work, with a default spending cap. |
| 28 | HubSpot | 500 to 5,000 credits included by edition; extra $10 per 1,000 | Shared credit pool | Its Customer Agent costs 50 credits per resolved conversation. One shared pool can be drained by one agent. |
| 29 | Salesforce Agentforce | Flex Credits $500 per 100,000 ($0.10 per standard action), plus editions and fixed-price deals | Credits per action, and more | Several meters at once shows buyers still want predictability. Keep a fixed-price option for big customers. |
| 30 | Clay | Launch from $185 a month: 2,500 Data Credits and 15,000 Actions | Two meters: data and platform work | Split what you resell from what you do, so you still earn when customers bring their own keys. |
Model #6: Per unit of work
One clear unit, one published rate. This is where infrastructure lives. It's the purest form of the usage-based pricing in Chapter Four of the book.
| # | Company | What they charge | Value metric | The lesson |
|---|---|---|---|---|
| 31 | Deepgram | Nova-3 pre-recorded $0.0043 a minute; voice agent $0.075 a minute | Audio minute | It discounts the voice agent when customers bring their own LLM. Let buyers swap parts and still buy from you. |
| 32 | Twilio | AI voice (ConversationRelay) from $0.07 a minute | Minute | AI voice priced like a phone call, a unit buyers already budget for. |
| 33 | Exa | Search $4 to $12 per 1,000 requests; agent effort levels $0.012 to $1.00 per request | Request, by depth and effort | Letting the caller pick an effort level puts a known price on agent work. |
| 34 | turbopuffer | Usage-based, with monthly minimums of $16, $256 or $4,096 | GB stored and written, PB queried | A minimum that counts toward usage sets a revenue floor without a separate platform fee. |
| 35 | Midjourney | $10 to $120 a month | Fast GPU hours, plus unlimited Relax mode | Meter your real cost (GPU time), but give a slow, unlimited safety net so nobody hits a wall. |
Which of these models fits your business?
Answer a few questions about your product, costs and buyers. The free Pricing Model Picker recommends flat, seat, usage, credits, outcome or hybrid.
Model #7: Per outcome
You pay only when the job is done. The easiest AI price to sell, and the hardest to define. See outcome-based pricing examples for more.
| # | Company | What they charge | Value metric | The lesson |
|---|---|---|---|---|
| 36 | Fin (formerly Intercom) | $0.99 per outcome; $1.99 per voice outcome; $9.99 per sales-qualification outcome | Outcome | Publish exactly what counts and refund reopened cases. According to MarTech and Pricing Innovation, Salesforce bought Fin for about $3.6 billion, closing 10 Sep 2026. |
| 37 | Zendesk | $1.50 committed or $2.00 pay-as-you-go per automated resolution, beyond 5 to 10 included per agent | Resolution, on top of seats | It added "dual verification" of billed resolutions. Build an auditable definition from the start. |
| 38 | Sierra | Reported about $1.50 per resolution; contracts reported from about $150,000 a year | Resolution or business outcome | Outcome pricing can fund fast enterprise growth when the outcome (say, a saved cancellation) is valuable. |
| 39 | Decagon | Per conversation or per resolution, customer's choice; reported platform fee about $50,000 a year | Conversation or resolution | Decagon says "the vast majority" of customers choose per-conversation. Offer a predictable alternative. |
| 40 | Paraform | Charges only when a hire is made; reported about 20 to 25% of first-year salary | Successful hire | Pure success fees work when the outcome is big, rare and unambiguous. |
What do the 40 examples have in common?
Step back from the tables and five patterns jump out.
1. Almost nobody charges a pure flat fee for heavy AI. Every company whose AI does expensive, variable work meters it somehow: an allowance, credits, minutes or outcomes. The flat-fee holdouts either own a huge workflow (Microsoft, Shopify) or sell against costly labor (Hebbia, and Harvey, reported at $1,200 to $2,000-plus per seat per month).
2. The hybrid has won. Eight of the 40 sell a seat or plan with an included allowance, and most of the credit companies wrap their credits in a monthly plan too. That matches the wider data: among the 2026 class of fast growers with visible pricing, 18 of 33 charge mainly on consumption, and only four mainly on seats. See how 2026's fastest-growing companies price.
3. Transparency is a competitive weapon. Lovable's example prompts, Copilot Studio's published rate card and Fin's exact outcome definition all reduce the fear of buying. The backlash cases (GitHub, Figma, Canva) were about changing what customers got without clear numbers.
4. Safety nets cut churn. Midjourney's Relax mode, Adobe's unlimited standard generations, ElevenLabs' rollover and Supabase's default spend cap all stop customers hitting a wall. A wall is where cancellations happen.
5. Minimum commitments are everywhere. turbopuffer's monthly minimums, Claude Enterprise's 20-seat minimum, Sierra's reported contract sizes and Salesforce's prepaid credit packs are all the book's minimum commitment model, reborn for AI.
My take
If you run a normal business adding AI to a product or service, the closest examples are Notion, Atlassian and MYOB, not OpenAI. Keep your existing price, include a sensible AI allowance, and charge separately only for the AI that does one specific, valuable job. Save the clever meters for when you know your usage data.
Which example should you copy?
| If you are... | Look at | Why |
|---|---|---|
| A SaaS company adding AI features | Notion (#16), Atlassian (#17), Slack (#4) | Bundle AI into a tier and use it to justify an upgrade or a modest price rise |
| An AI app for creators or prosumers | Lovable (#23), ElevenLabs (#24), Runway (#25) | Credits with a published rate card and rollover |
| A developer tool or API | Deepgram (#31), Exa (#33), turbopuffer (#34) | Per-unit rates, a free credit to start, and a monthly minimum |
| An AI support or sales agent | Fin (#36), Zendesk (#37), Decagon (#39) | Outcome pricing with a predictable per-conversation option |
| A service firm (accountants, agencies, lawyers) | MYOB (#8), Hebbia (#5), Paraform (#40) | Price the job or the outcome, anchored on the labor it replaces |
| Selling to large enterprises | Claude Enterprise (#12), Salesforce (#29) | Seat plus pooled usage, with a fixed-price option for procurement |
Still not sure? The Pricing Model Picker will narrow it down in a few minutes.
Pricing Model PickerFlat, tiered, seat, usage, credits, outcome or hybrid? Find out.Open the free tool →What to do next
- Pick two or three examples from the table above that look most like your business.
- Read their teardowns in the pricing teardowns library and note their value metric, allowance and overage rules.
- Cost your own product with the method in the AI pricing guide.
- Borrow one transparency move (a rate card, an example cost table or a clear outcome definition) and one safety net (rollover, a slow mode or a default cap).
- Write down what you'll charge and test it on a handful of real customers before you publish it.
And remember the line from the book I repeated on purpose: you don't have to stick to one model. The best companies on this list don't.
Frequently asked questions
What is the most common AI pricing model in 2026?
A hybrid: a subscription or seat that includes a usage allowance, with credits or overage beyond it. ICONIQ's 2026 survey found AI companies blend an average of 1.7 pricing models, and consumption pricing rose from 35% to 42% of companies. Pure per-seat pricing is now the exception for AI-native products.
Which AI companies use outcome-based pricing?
Mostly customer-support and recruiting tools. Fin charges $0.99 per outcome, Zendesk $1.50 to $2.00 per automated resolution, HubSpot $0.50 per resolved conversation and Salesforce $2 per resolution for its Help Agent. Sierra and Decagon sell outcome pricing in custom contracts, and Paraform charges only when a hire is made.
Do any AI companies still offer unlimited plans?
Rarely for expensive features. Midjourney offers unlimited Relax mode on higher plans, and Adobe keeps standard generations unlimited after premium credits run out. These work because the unlimited part is slower or cheaper to run. Cursor and HeyGen both moved away from 'unlimited' once heavy users cost too much.
Are these AI prices in US or Australian dollars?
US dollars unless marked A$. Xero and MYOB are shown in Australian dollars because they publish Australian pricing. Most AI vendors bill Australian customers in US dollars, so the price you actually pay moves with the exchange rate. Australian businesses should also check how GST applies to each purchase.
How were these 40 examples chosen?
They come from our AI pricing dataset of 75 companies, checked on 27 October 2026, plus the 2026 class of fast growers that crossed about $100M ARR. We picked companies with published prices or well-sourced estimates, and at least one clear lesson for other businesses. Estimates from third parties are labeled 'reported.'
Get both books free
The Subscription Playbook (ebook + audiobook) and Feed a Starving Crowd, instantly. In Australia? Claim the free hardcover.
Robert Coorey, MBA
Robert is the author of The Subscription Playbook and Feed a Starving Crowd and a Director at BPAG Accountants in Sydney. More about Robert.
Sources
- SaaStr: 10 metrics from ICONIQ's 2026 State of AI report (10 Jul 2026)
- Microsoft: Microsoft 365 Copilot pricing (checked 27 Oct 2026)
- Claude: pricing (checked 27 Oct 2026)
- GitHub: Copilot plans (checked 27 Oct 2026)
- Lovable: pricing (checked 27 Oct 2026)
- Intercom (Fin): pricing (checked 27 Oct 2026)
- Zendesk: pricing (checked 27 Oct 2026)
- Salesforce: Agentforce pricing (checked 27 Oct 2026)
- Value Add VC: How Sierra AI makes money (updated 21 Oct 2026)
- Xero Australia: pricing plans (checked 27 Oct 2026)
- MYOB Australia: pricing (checked 27 Oct 2026)
- Deepgram: pricing (checked 27 Oct 2026)
Facts and prices checked 27 October 2026. Spotted something out of date? Tell us.