Someone in your last budget meeting asked what the AI agents in your CRM are going to cost. If the answer was a shrug, you are in good company. Zoho Zia Agents pricing is unusual enough that plenty of teams who have already switched the agents on still cannot explain their own bill.
The short version: the agent builder is free. The thinking is not. Zoho charges for language model usage out of a prepaid wallet, hands you a monthly allowance of free tokens, then bills the overage per million tokens. That is a fairer model than the per-user AI add-on most vendors sell, because a team of forty that barely uses agents pays close to nothing. It is also much harder to forecast, which matters a lot if you are the person signing the number.
What Zia Agents actually are
An agent here is not a chatbot bolted to the corner of a screen. It is a configured worker: you give it a purpose, written instructions, a knowledge base to read from, and a set of tools and connections it is allowed to use. It then runs against your CRM data on its own.
Zoho ships seven pre-built agents for CRM that you can hire from its Agent Store rather than build: a Sales Development Representative that nurtures and qualifies inbound leads, a Sales Coach that runs role-plays and gives reps feedback, a Follow-up Scheduler, a Deal Analyzer that estimates win probability, a Quote Generator, a Revenue Growth Specialist for upsell and cross-sell, and a Deal Closure Reminder that sends a summary three days before an expected close date. Agent Studio is where you build your own.
Two details from the documentation are worth having before anyone books a demo.
Edition and region. Zia Agents are available in Professional and above, and only in the US, India, EU and Australia data centres. Zoho runs data centres in Canada, Japan and China too, and those are not on the list. If your org was provisioned outside those four regions, the feature is not something you can turn on, whatever edition you are paying for. Setting one up also requires the Zia Agents Manage Configuration permission, which your admin controls.
How the agent signs its work. You can deploy an agent in one of two ways. Via a connection, the agent acts under the identity of whoever deployed it, so every email and record change lands in the audit trail under a real person's name. Or as a Digital Employee, where the agent gets its own entity, role and email address. The second option costs a little more setup and is worth it, because six months from now somebody will ask who sent a particular email to a customer, and "the VP of Sales, apparently" is not the answer you want.
How much do Zoho Zia agents actually cost?
The platform itself is free to use. You pay only for the model that does the reasoning, out of a prepaid AI Credits wallet where 1 USD buys 1,000 credits. Credits roll over month to month. Only a super-admin can buy them, through the Zoho Store, and the checkout takes USD or INR.
| Model option | Free tokens per month | Cost after that |
|---|---|---|
| Standard tier (Qwen 14B, Qwen 3.5 35B MoE, GLM 4.7 Flash) | 30 million | 1.00 USD per 1M tokens |
| Pro tier (GLM 5, US and India data centres only) | 20 million | 3.00 USD per 1M tokens |
| OpenAI models through Zoho's keys (GPT-4o-mini up to GPT-5.4) | None | Credits deducted from the first token |
| Your own API key (OpenAI, Anthropic, Gemini, DeepSeek, Cohere) | Not applicable | Your vendor bills you, with no Zoho markup |
Read that third row twice. It is the line that catches people. Zoho's own models come with a large free allowance; the brand-name OpenAI models offered in the same dropdown come with none, and start drawing down your wallet immediately. An admin choosing a model by which name they recognise, rather than by what it costs, is how a free feature becomes a line item.
What this means for your business
For a sales team of twelve running two or three agents on routine work, the realistic outcome is that you stay inside the 30 million free tokens on the Standard tier and pay nothing at all. That is the honest read, and it is a genuinely good deal compared with paying per seat for AI whether your reps use it or not.
Where it stops being free is volume and context. Token consumption does not scale with how many people you employ; it scales with how much text the agent reads and writes each time it runs. An agent with a large attached knowledge base re-reads that material on every invocation. An SDR agent handling 200 inbound leads a month with a multi-turn email exchange each consumes far more than a Deal Closure Reminder that fires a short summary a few times a week.
Zoho does not publish a tokens-per-task figure, so anybody who tells you what an SDR agent costs per lead is guessing. What you can do instead is measure it. The settings area breaks usage down by model and by agent, over a period you choose, so after four weeks of real running you will have your own number rather than a vendor's. Budget the first month as an experiment, not as a forecast.
Separate the strategic case from the cost case. Zoho's published 24-month AI roadmap puts its heaviest investment into autonomous sales orchestration and buyer intelligence, and explicitly reduces investment in complex rule-based workflows with no AI or buyer-signal component. Existing workflows keep running. But if your CRM automation is a large pile of hand-built rules, the platform is walking in a different direction from you.
The catch
A free platform with a metered engine has a specific failure mode, and it is worth naming.
- There is no documented spend alert or hard cap. The usage dashboard is a rear-view mirror. If an agent gets stuck in a loop or someone attaches a 200-page knowledge base, you find out in the report, not at the moment it happens. Check your wallet weekly for the first month.
- Only super-admins can top it up. Zoho does not document what happens to a running agent when the wallet empties. If your one super-admin is on holiday, that is an operational question you want answered before an agent is doing customer-facing work, not after.
- GLM 5 is US and India only. If you are on the EU or Australia data centre and you were planning on the Pro tier, check before you plan around it.
- The agents read your CRM as it is. A Deal Analyzer trained on a pipeline where half the opportunities have a stale close date will produce confident nonsense. Nothing about the pricing model changes the oldest rule in CRM: the data work comes first, and it is the part nobody budgets.
- Model choice is a cost decision made by a technical person. The dropdown that picks your model has a 3x price spread across Zoho's own tiers and an infinite one against external models. That decision should not be made casually.
Sizing the token budget and picking the right model for each agent is the kind of thing worth half an hour with someone who has watched it go wrong, before you commit an agent to customer-facing work.
What to do next
Four steps, in order, and you can genuinely do the first two yourself this week.
- Confirm your edition and data centre. Professional or above, in US, India, EU or Australia. If either is wrong, stop here and price the upgrade before anything else.
- Hire one pre-built agent, not three. The Deal Closure Reminder or the Follow-up Scheduler are low-risk starts because neither writes to a customer unsupervised.
- Run it on Standard tier for four weeks, then open the usage breakdown by agent. That report is your actual cost model. Everything before it is arithmetic.
- Decide on the deployment mode before an agent emails a customer. Digital Employee, with its own identity and audit trail, is the answer for anything customer-facing.
The teams that get value from this are not the ones that switch on all seven agents in a week. They are the ones that pick the single most repetitive thing their reps do and hand over that one job properly.
Get a straight answer on your own numbers
If you want to know what Zia Agents would actually cost in your org, and whether your edition and data centre even support them, we run a free 30-minute Zoho setup and licence review. We look at your current edition, your user count, the automation you already have running and the agents that would realistically pay for themselves, and send you a one-page summary: what you can turn on today, what needs a licence change, and what needs data cleanup first.
Zoho is shifting its development investment toward agent-driven orchestration over the next two years, which means this stops being an experiment and starts being a budget line at your next renewal. Better to have the number before that conversation than during it. Book a review with DuCodes.
Sources
Written from Zoho's own documentation: Zia Agents pricing, model configuration, platform settings, Zia Agents in Zoho CRM and the Zoho CRM AI roadmap. Prices and allowances were accurate on 13 September 2026 and Zoho changes them without a release cycle, so re-check before you budget.
