Researched
This analysis is based on documentation, public user reports, and vendor materials — not yet on our own hands-on testing. How we rate
First, the naming, because it will confuse you
HubSpot's AI layer was called Breeze. At the end of July 2026 it was renamed Agent Hub, at the same time HubSpot added Revenue Hub as a sixth product line alongside Marketing, Sales, Service, Content and Data.
Most documentation still says Breeze. Most third-party coverage still says Breeze. HubSpot's own materials are inconsistent. You'll encounter both names referring to the same set of capabilities: an in-app assistant, a family of autonomous agents, and data enrichment.
Nothing about the functionality changed with the rename. Only the label.
The pricing is three layers deep
Somebody on a CRM forum described HubSpot's pricing as more of a pyramid than a price list, which is uncharitable and accurate.
Layer one: seats. HubSpot moved to a seats-based model where each tier bundles a baseline number of Core Seats — one on Starter, three on Professional, five on Enterprise — with additional seats billed monthly on top. Sales Hub runs roughly $15 Starter, $100 Professional, $150 Enterprise per seat per month annually. Service Hub sits at about $20, $90 and $150. Marketing Hub Professional starts considerably higher, around $800 a month, and the full Customer Platform Professional bundle starts around $1,300.
Layer two: marketing contacts. If Marketing Hub is in your stack, you also pay by the number of contacts you actively market to, in blocks of a thousand. This scales independently of seats and independently of AI usage.
Layer three: credits. Every AI feature meters against a pool of HubSpot Credits. Each paid tier bundles a monthly allowance — 500 on Starter, 3,000 on Professional and 5,000 on Enterprise for most hubs, with Data Hub and Customer Platform bundles running 500, 5,000 and 10,000. Additional credits cost around $10 per 1,000, or closer to $9 with an annual commitment.
Then there's onboarding, which is mandatory on professional tiers and doesn't appear on the calculator: roughly $1,500 for Sales or Service Professional, around $3,000 for Marketing Professional.
Outcome-based pricing is the genuinely interesting change
In April 2026 HubSpot did something most vendors haven't: it started charging for results rather than attempts.
| Agent | Cost | Billed on |
|---|
| Customer Agent | 50 credits (~$0.50) | Per resolved conversation |
| Prospecting Agent | 100 credits (~$1.00) | Per recommended lead |
| Data Agent | 10 credits (~$0.10) | Per answer |
Previously the Customer Agent charged per conversation regardless of whether it resolved anything, and the Prospecting Agent carried a recurring per-contact fee. The shift means a failed attempt costs you nothing, which is a meaningful alignment of incentives and rare enough that it's worth noting HubSpot as one of the larger incumbents doing it.
The arithmetic still gets serious at volume, though. Fifty cents per resolution sounds trivial until you're a support operation handling ten thousand conversations a month, at which point the agent alone is running into four figures on top of your Service Hub seats. Do that calculation with your actual ticket volume before assuming outcome pricing makes this cheap.
The auto-upgrade setting nobody mentions
Here's the detail most likely to produce an unpleasant invoice.
If your credit consumption exceeds your monthly allowance, HubSpot automatically upgrades your account to a larger credit tier — unless you've explicitly switched on pay-as-you-go billing instead, which invoices overage in small blocks at the standard rate.
Auto-upgrade is the default behavior. It's documented, it's not hidden, and it will still catch people who assumed hitting a limit meant features stopped working rather than the plan getting bigger.
Check this setting the day you enable AI features. Decide deliberately which behavior you want. Pay-as-you-go gives you predictable marginal cost; auto-upgrade gives you uninterrupted service and a larger bill.
The agents are only as good as your CRM
This is the thing practitioners say most consistently and it deserves more weight than the feature list.
HubSpot's agents reason over what's in HubSpot — records, tickets, conversation history, knowledge base articles. They inherit HubSpot's permission model, which is a genuine architectural advantage over bolting a general AI tool onto your CRM. What they can't do is know things your CRM doesn't.
Where data is clean and complete, output is genuinely useful. HubSpot publishes reference customers with strong numbers — high chat deflection rates, meaningful lead generation lift — and reports that teams using the Customer Agent close substantially more tickets monthly. Those are vendor-reported and directionally plausible for organizations with mature data.
Where data is patchy, the agents produce confident answers built on gaps. The most common criticism isn't model quality. It's that the inputs are the weak point, and no amount of AI fixes a CRM nobody maintains.
The related limitation: the Customer Agent leans heavily on HubSpot knowledge base articles. If your actual support documentation lives in Notion, Confluence or an external helpdesk, you're migrating it or accepting worse answers. There are workarounds and none of them are enjoyable.
How we researched this
We don't run HubSpot at a tier where the AI features are meaningfully exercised, and this page reflects research rather than operational experience.
The basis: HubSpot's published pricing and product catalog documentation for seat rates, credit allowances and overage behavior, cross-referenced against independent pricing analyses published between May and August 2026 — including several that captured figures directly from HubSpot's live pricing page, one as recently as late June. Product changes and the Agent Hub rename come from HubSpot's own announcements. Adoption and outcome statistics are HubSpot's and flagged as such.
The main research difficulty is that HubSpot's pricing has many moving parts and third-party sources tend to cover one layer well and gloss over the others. Seat prices are widely reported and reliable. Credit allowances are documented but rarely mentioned in reviews. Onboarding fees appear in perhaps a third of the coverage. We've tried to assemble the full picture, but the only number that will be accurate for your situation comes from HubSpot's calculator with your actual seat count and contact volume plugged in.
What research can't tell you: whether the agents perform well on your data. That's a pilot, and given the data-quality dependency, it's a pilot worth running before committing.
Where this doesn't work
Your knowledge lives elsewhere. Covered above, and it's the most common structural mismatch.
You're shopping for AI, not a CRM. HubSpot's AI is a reason to stay on HubSpot. It's a poor reason to migrate to it. Choose the CRM on CRM merits.
Very high support volume. Outcome pricing is fair per unit and still adds up. Run the numbers against your ticket count before assuming it's cheap.
Small teams on Starter. Five hundred credits a month is not much once agents are doing real work, and Starter locks you out of the flagship agents anyway — the Customer Agent requires Professional or above.
Budget predictability is critical. Between seat counts, contact tiers, credit consumption and auto-upgrade, forecasting a HubSpot bill twelve months out is genuinely hard.
Against the alternatives
Against Salesforce Einstein: the same structural argument on the other platform. Einstein is a strong reason to stay on Salesforce. HubSpot remains meaningfully more accessible for SMB and mid-market buyers on both price and setup complexity; Salesforce wins on enterprise depth and customization.
Against Gong: Gong does conversation intelligence far more deeply, and costs far more. HubSpot's AI is broad and shallow across the whole funnel; Gong is narrow and deep on revenue conversations. Organizations at scale run both.
Against Clay: Clay's enrichment is substantially more powerful than HubSpot's built-in data features, and Clay typically feeds HubSpot rather than competing with it.
Against Intercom, Sierra AI and Decagon: all are more focused on AI customer support specifically and generally better at it. HubSpot's Customer Agent wins when the deciding factor is native CRM context and permissions; the specialists win on resolution quality and knowledge flexibility.
Against Zapier or Make plus a general assistant: you can assemble something similar for less money, and you'll spend engineering time and lose the permission inheritance. For teams already deep in HubSpot, the native option is usually worth the premium.
Pricing 2026
| Component | Cost | Notes |
|---|
| CRM (free tier) | $0 | Core CRM with limited AI |
| Sales Hub | ~$15 / $100 / $150 per seat/mo | Starter / Professional / Enterprise, annual |
| Service Hub | ~$20 / $90 / $150 per seat/mo | Same tier structure |
| Marketing Hub Professional | From ~$800/mo | Plus marketing contact tiers |
| Customer Platform Professional | From ~$1,300/mo | Bundles multiple hubs |
| Credit allowance | 500 / 3,000 / 5,000 per month | Starter / Pro / Enterprise, most hubs |
| Extra credits | $10 per 1,000 ($9 annual) | Auto-upgrade unless PAYG enabled |
| Customer Agent | ~$0.50 per resolved conversation | Professional and above |
| Prospecting Agent | ~$1.00 per recommended lead | Starter and above |
| Data Agent | ~$0.10 per answer | Starter and above |
| Onboarding | ~$1,500–3,000 one-time | Mandatory on professional tiers |
Checked August 2026. HubSpot renamed Breeze to Agent Hub at the end of July 2026 and added Revenue Hub as a sixth product line. Flagship agents moved to outcome-based pricing in April 2026. Seat prices vary by hub and billing period; figures above reflect annual billing. Credit allowances and overage behavior are documented in HubSpot's product catalog. Use HubSpot's own calculator with your seat count and contact volume for an accurate figure.
Check the auto-upgrade setting immediately. Decide between automatic tier increases and pay-as-you-go before you start consuming credits, not after.
Add onboarding to your first-year budget. It's mandatory on professional tiers and absent from the calculator.
Clean your CRM before evaluating the AI. The agents are downstream of your data quality. A pilot on messy records tells you nothing useful about the product.
Model your support volume against outcome pricing. Fifty cents a resolution is fair and it compounds. Ten thousand conversations a month is a real number.
Our Verdict
HubSpot's AI is well-executed and correctly positioned. Building agents that inherit the CRM's permission model and reason over records already in the system is architecturally sensible, and the outcome-based pricing move is genuinely buyer-friendly in a market where most vendors bill for attempts.
It's a retention feature, not an acquisition one. The value is entirely a function of your data already living in HubSpot. Nobody should migrate CRMs for this, and HubSpot's AI shouldn't tip a platform evaluation that's otherwise close.
The credit layer is the part to plan around. Between allowances, overage rates, auto-upgrade defaults and outcome-based agent charges, this is not a pricing model you can hold in your head. Model it properly with real volumes before committing.
Data quality is the actual dependency. Every honest assessment of HubSpot's agents converges on the same point: the models are fine and the inputs are usually the problem. If your CRM is a mess, the AI will confidently reflect that mess back at you.
For a company already committed to HubSpot with reasonably clean data: this is a solid addition and the outcome pricing makes it easier to justify than it was a year ago. For a company shopping for AI capabilities: choose your CRM on CRM merits and treat the AI as a bonus. For organizations whose knowledge lives outside HubSpot: the structural mismatch is real and a specialist tool will serve you better.
Note: AIVario earns no commission from HubSpot. This page is based on published documentation, independent pricing research and HubSpot's own announcements rather than a production deployment at the relevant tier.
Best for: Existing HubSpot customers with clean CRM data, mid-market teams wanting AI inside their system of record, support operations whose knowledge base already lives in HubSpot, organizations valuing native permission inheritance
Not ideal for: Companies choosing a CRM primarily on AI capability, organizations with knowledge scattered across external tools, very high-volume support operations, teams needing predictable monthly costs
Bottom line: Good AI in the right place, wrapped in pricing that takes real effort to model. Excellent if you're already here; not a reason to come here.
- Salesforce Einstein — the same argument on the other major CRM platform
- Gong — far deeper conversation intelligence at far higher cost
- Clay — enrichment that outclasses HubSpot's built-in data features
- Intercom — specialist AI customer support with more flexible knowledge sources
- Sierra AI — purpose-built AI support agents for high-volume operations
- Apollo.io — prospecting data that feeds the top of the HubSpot funnel
- Make — the assemble-it-yourself alternative to native automation
Frequently Asked Questions about HubSpot AI
How is HubSpot's AI priced?
Three layers stacked on each other. First, a per-seat hub subscription — Sales Hub runs roughly $15 Starter, $100 Professional and $150 Enterprise per seat per month on annual billing, with Service Hub at about $20, $90 and $150. Second, if you touch Marketing Hub, you pay by marketing contacts in blocks of a thousand. Third, HubSpot Credits meter every AI feature at roughly $10 per 1,000, with each paid tier bundling a monthly allowance. On top of all that, the flagship agents are now billed per outcome rather than per attempt.
What is outcome-based pricing and what does it cost?
HubSpot moved its main agents to charging for results rather than attempts in April 2026. The Customer Agent charges 50 credits — about $0.50 — per resolved conversation, replacing a previous flat per-conversation rate. The Prospecting Agent charges 100 credits, roughly $1.00, per recommended lead, replacing a recurring per-contact fee. The Data Agent charges 10 credits, about $0.10, per answer. The genuine advantage is that you don't burn budget on failed attempts, which is unusual and buyer-friendly. The catch is that a high-volume support operation can still run up serious numbers.
What are HubSpot Credits and how many do I get?
A single consumption pool that funds every AI feature across the platform. Baseline allowances for most hubs are 500 credits on Starter, 3,000 on Professional and 5,000 on Enterprise; Data Hub and Customer Platform bundles run 500, 5,000 and 10,000. Extra credits cost around $10 per 1,000 at the standard rate, closer to $9 with an annual commitment. Important detail: if you exceed your monthly allowance, HubSpot auto-upgrades your account to a larger tier unless you've explicitly enabled pay-as-you-go instead. That's a setting worth checking on day one.
Breeze or Agent Hub — which is it?
Agent Hub, as of the end of July 2026. Breeze was the previous name for HubSpot's AI layer and the rename happened alongside the addition of Revenue Hub as a sixth product line. Most documentation and third-party coverage still says Breeze, and HubSpot's own materials are mixed, so you'll encounter both. The capabilities are the same — an in-app assistant, a set of autonomous agents, and data enrichment. Only the branding changed.
Are there onboarding fees?
Yes, and they're mandatory on professional tiers. Reported one-time fees run around $1,500 for Sales or Service Professional and roughly $3,000 for Marketing Professional, with higher figures at Enterprise. These don't appear on the pricing calculator and they're a common budgeting surprise. Factor them in before comparing HubSpot's headline seat price against a competitor's, because the first-year total is meaningfully higher than the monthly rate suggests.
Is HubSpot's AI actually good?
It's good in proportion to how clean your CRM data is, which is the honest framing. The agents work by reasoning over what's in HubSpot — records, tickets, knowledge base articles, conversation history. Where that data is complete and well-maintained, the output is genuinely useful. Where it's patchy, the agents produce confident answers built on gaps. The most consistent criticism from practitioners isn't about the model quality; it's that the inputs are usually the weak point. Clean the CRM before you evaluate the AI.
What if our knowledge lives outside HubSpot?
That's the significant structural limitation. The Customer Agent draws primarily on HubSpot knowledge base articles, so if your actual support documentation lives in Notion, Confluence, Google Drive or an external helpdesk, you're either migrating it or accepting degraded answers. Workarounds exist and none are pleasant. For organizations whose institutional knowledge is scattered, a retrieval tool that indexes across sources will serve better than HubSpot's agent will.
Should I pick HubSpot for its AI?
No, and that's not a criticism of the AI. HubSpot's AI is a strong reason to stay on HubSpot and a weak reason to move to it. The value comes from the agents having native access to your CRM records, permissions and workflows — which only matters if that's where your data already is. If you're evaluating CRMs from scratch, decide on the CRM merits: usability, ecosystem, pricing at your scale, and how well it fits your sales motion. The AI layer is a tiebreaker at most.
