What AI support really costs
There are four reasonable paths and none is the best for everyone. Here they are side by side, with public prices, their sources and a calculator where you can change every assumption.
A helpdesk suite
Zendesk, Freshdesk or HubSpot: you pay per person who handles tickets, and the AI is charged per resolution or in credits.
An AI agent that resolves on its own
Intercom with Fin or Intake: you pay per resolved conversation or by tier, and the AI answers without anyone stepping in.
Build it with frontier models
With Claude Code and the Claude, GPT or Gemini API you build your own agents. You pay for hours and tokens.
Elevatia
Support, accounts and knowledge in one place. The AI drafts with the account context and a person approves.
List prices excluding VAT, checked on Oct 16, 2026 on each company's website. Brands belong to their owners; Elevatia has no relationship with them. If you spot outdated data, write to us and we will fix it.
Total cost, with your numbers
Move the controls. Each platform is calculated on its cheapest plan that already includes an AI agent, billed annually and with no add-ons: if one comes out expensive, it is not because we padded it with extras.
Assumptions (editable)
They mostly affect "Build it with Claude Code". Tokens per conversation are an estimate until we publish the measurement.
With these numbers, building it with Claude Code does not cost less than Elevatia in five years: the build and maintenance weigh more than the license.
Does not include the salaries of the support team, which are the same in every option. Dollar prices converted at €1 = $1.1253 (ECB, Oct 16, 2026). Elevatia's price is the planned price after launch; today it is free.
Tokens are no longer the expensive part
This is what a support conversation costs in tokens with each model, using the same usage estimate for all of them. It is what you would pay the API if you build it yourself; on the platforms it is included in the price.
| Model | Provider | Input / output ($ per million) | € per conversation | € per 1,000 conversations |
|---|---|---|---|---|
| Claude Opus 5.5 | Anthropic10 | 4 / 20 | €0.0746 | €74.65 |
| Claude Sonnet 5.5 | Anthropic10 | 2 / 10 | €0.0373 | €37.32 |
| Claude Haiku 5.5 | Anthropic10 | 0.1 / 0.5 | €0.0019 | €1.87 |
| GPT-6 Astra | OpenAI12 | 10 / 50 | €0.1866 | €186.62 |
| GPT-6.1 Sol | OpenAI12 | 2 / 10 | €0.0373 | €37.32 |
| GPT-6 Luna | OpenAI12 | 0.1 / 0.5 | €0.0019 | €1.87 |
| Gemini 4 Argon · launch price until Nov 30, 2026 | Google13 | 2 / 10 | €0.0373 | €37.32 |
| Gemini 3.1 Pro | Google13 | 2 / 12 | €0.0398 | €39.81 |
| Gemini 3.8 Flash · launch price until Dec 31, 2026 | Google13 | 0.75 / 3.75 | €0.0140 | €14.00 |
| gemma4 (31B) · the one Elevatia uses | Ollama Cloud14 | 0.14 / 0.4 | €0.0022 | €2.24 |
| glm-5.3-flash · the one Elevatia uses | Ollama Cloud14 | 0.15 / 0.5 | €0.0025 | €2.49 |
Per token is not the same as per resolution
Intercom charges $0.99 per conversation that Fin resolves and Zendesk €1.50 per resolution above the included ones. That price pays for the product, the infrastructure, the testing and the risk, not just the tokens. Comparing $0.99 with a few cents of API is unfair to them: what really costs money is everything else.
What does change
A small model from Anthropic, OpenAI or Google already costs the same as or less than the open models we use. The large frontier models reason better than ours on hard problems; the price difference is real, but for a support reply it is rarely what decides. What decides is who builds, tests and maintains everything around the model.
With Claude Code you build an agent map in days. And then?
It is true: with Claude Code, subagents, MCP and a frontier model API, one technical person can build in a few days agents that read tickets, search your documentation and draft replies. It is a serious option, and in many cases the right one. This section covers what sits between that first version and something your team uses every day.
What you gain by building it yourself
The best model of the moment
You choose Claude, GPT or Gemini and switch when a better one comes out. A reasoning quality that an open model does not offer today.
Total control
Your code, your data, your rules. It adapts to the millimeter to a workflow of your own that no product covers.
Speed at the start
A useful prototype in days, with no contracts or migrations. Ideal for validating whether AI really helps your support.
What is behind Elevatia and you would have to build
Hours of an experienced person using Claude Code, which speeds things up a lot. They do not include product design or the back-and-forth with your team, which often costs as much as building. You can change them in the calculator.
| Inbound channels | Connect Gmail and Outlook through OAuth, rebuild threads, attachments, replies from the platform. | 60 h |
|---|---|---|
| Tickets and SLA | Statuses, assignment, priorities and deadlines with a business calendar and holidays, alerts before a breach. | 50 h |
| Companies, users and permissions | Several isolated organizations, roles, per-action permissions, sign-in with Google and Microsoft. | 50 h |
| AI agent | Triage that applies priority and SLA, drafts with the account context, tools with permissions. | 60 h |
| Knowledge base | Import documents, chunk them, semantic search, articles published for customers. | 40 h |
| Approval and audit trail | An inbox where a person approves what the AI proposes, and a log of who decided what. | 30 h |
| Accounts and contacts | Companies, contacts, opportunities and a record with everything that has happened with each account. | 50 h |
| Connectors | Import from Zendesk and Intercom, read Stripe and Odoo, signed webhooks, MCP servers. | 50 h |
| Quotas and cost | Usage limits per company so a traffic spike does not turn into a token bill. | 20 h |
| AI testing | Test cases that warn you when a prompt or model change makes the answers worse. | 30 h |
| Deployment and operations | Servers, backups, monitoring, security updates. | 30 h |
| Total | ≈ €21,150 at 45 €/h | 470 h |
Costs that do not show up on the token bill
Models change
Providers retire versions and change their behavior. Every change calls for reviewing prompts and rerunning the tests.
Usage spikes
A poorly bounded agent loop or a customer who writes a lot can multiply the bill overnight if there are no limits.
Data protection
Tickets carry your customers' personal data. You need to sign the DPA with the model provider and document where it is processed.
Who maintains it
When the person who built it moves to another project, the system is left without an owner. That risk has no price in the calculator.
When building it yourself is the right answer
- You have a technical team with time and you want it to be part of your product.
- Your workflow is so specific that no platform covers it without forcing it.
- Your volume is high and steady, and in the calculator the break-even point falls within your horizon.
- You need the best reasoning quality in every reply and the cost per token does not matter to you.
It is not "one or the other"
We are working on letting your Claude Code or your Claude connect to Elevatia through MCP: your agents read tickets, accounts and knowledge with your permissions, and whatever they want to change lands in an inbox for someone to approve. Today it is built but not published; we will tell you when it is open.
Point by point, no winner marks
Each cell says what that option does, not whether it does it "well". When a data point could not be verified on the official website, we say so. The numbers next to each row lead to its source.
| ElevatiaLaunching | ZendeskService suite | IntercomHelpdesk + Fin | FreshdeskService suite | HubSpotService Hub | IntakeAI agent by tier | Claude CodeBuild it yourself | |
|---|---|---|---|---|---|---|---|
| Price | |||||||
| How it is charged13568 | By tiers of conversations per month, with unlimited users. Free during the launch. | Per agent per month, plus AI resolutions above the included ones. | Per seat per month, plus $0.99 per conversation that Fin resolves. | Per agent per month, plus AI agent sessions and per-agent copilot as extras. | Per seat per month, plus AI credits that are spent on each resolution. | By tiers of conversations per month, with unlimited users. | Build and maintenance hours, Claude Code licenses and API tokens. |
| Starting at136811 | Planned: €49/month for 300 conversations; €199 for 1,500; €449 for 5,000. | €19/agent/month (Support Team, annual). With AI agents, from €55 (Suite Team). | $19/seat/month (Essential, annual), $85 Advanced, $132 Expert. | €17/agent/month (Growth, annual), €52 Pro, €84 Enterprise. | There is a free plan. Professional: $100/seat/month according to its catalog; the euro price could not be read. | €49/month for 200 conversations; €129 for 800; €289 for 2,000 (annual). | Claude Code from $20/month; for a team, a premium seat at $125/month. Plus tokens and hours. |
| AI included16 | All the AI is inside the tier; there is no charge per resolution or per token. | 5 resolutions per agent per month (10 in Professional); after that, €1.50–2.00 each. | None included: every Fin resolution is paid. There are configurable spending limits. | 500 AI agent sessions, one time per account; after that, €45 per 100 sessions. | 3,000 credits per month in Professional; the agent spends 50 per resolved conversation. | Inside the tier. Once the tier is used up, the agent stops until next month; there is no surcharge. | You pay every token to the API. No limits except the ones you program. |
| Copilot for the human agent13 | Included: drafts with the account context for a person to review. | Add-on at €50/agent/month. | Add-on at $29/agent/month (annual). | Freddy Copilot: €29/agent/month (annual), only in Pro and Enterprise. | Breeze Assistant is included in the seat. | The focus is on the agent resolving on its own; there is handoff to a human with context. | Whatever you build. |
| Trial138 | Free during the launch, no card. | 14 days of Professional. Startup program with up to two years free. | 14 days with no card. Startup program with up to 93% off. | 14 days on the Enterprise plan, no card. | Permanent free plan with basic features. | 10 days with no card, no commitment. | You can start today with a Pro account. |
| Channels and tickets | |||||||
| Channels138 | Email (Gmail and Outlook) and customer portal. WhatsApp, coming. | Email, messaging, chat, voice, social and WhatsApp (some at an extra fee). | Chat, email, phone, Slack, social, WhatsApp and SMS. | Email, portal and web widget. Chat, phone and messaging in Freshdesk Omni (separate product). | Shared inbox, chat, chatbot and calls. | Email, WhatsApp and web chat in the same inbox. | The ones you connect. |
| SLA31 | Included, with a business calendar; the AI assigns it when classifying the ticket. | Available; which plan it comes in could not be verified. | Only in the Expert plan. | Rules in all plans; multiple SLA policies from Pro. | Yes, SLA management. | Not published. | Whatever you build. |
| Portal for your customers | Yes: your customers see their tickets and the published knowledge base. | Could not be verified on the public website. | Help center; a per-customer ticket portal could not be verified. | Yes, in all plans; with your branding from Pro. | Yes, customer portal. | Public help center; ticket portal not published. | Whatever you build. |
| Voice and contact center13 | No. | Yes: telephony, IVR and contact center as an add-on. | Phone and Fin Voice ($1.99 per voice resolution). | In Freshdesk Omni, a separate product. | Calls included in Professional. | Not published. | Possible with voice APIs; separate work. |
| What the AI does | |||||||
| Approach | Proposes and a person approves: it drafts, classifies and suggests actions with the account context. | AI agents that resolve on their own across several channels, plus a copilot for agents. | Fin resolves as much as possible on its own; they publish a 76% average resolution rate. | Freddy AI Agent answers on its own by email and web; Copilot helps the agent from Pro. | Customer Agent that resolves and an assistant for the team, on top of the CRM. | The agent resolves on its own, even performing actions against your product's API. | The one you design: autonomous, with approval or mixed. |
| Actions with human permission | Yes: anything sensitive goes to a "To approve" inbox and who approved is recorded. | Action builder from Suite Team; approval details not checked. | Fin runs workflows and procedures; approval details not checked. | MCP actions included per year depending on the plan; approvals not checked. | Not checked. | Calls your API with the permissions you give it. | Claude Code asks for permission per tool; in production, whatever you program. |
| Which model it uses24914 | Open models (gemma4 and glm-5.3-flash) on Ollama Cloud, chosen per turn depending on difficulty. | OpenAI with zero retention or models on Azure, Bedrock or Google Cloud. | Own model (Fin Apex) plus unnamed external providers. | Not published on the pages we read. | Not checked. | Models on Amazon Bedrock in EU regions; specific model not published. | The one you choose: Claude, GPT, Gemini or an open one. |
| Trains on your data492 | No. | They state they do not train on customer data. | May use anonymized data for fine-tuning; it can be turned off. | Not checked. | Not checked. | They state they do not train on your data. | Depends on the terms of the API you sign up for (by default, they do not train). |
| Accounts and knowledge | |||||||
| CRM | Included: companies, contacts, opportunities, agenda and a 360 record per account. | It has a separate CRM product (Sell); whether it is still on sale could not be verified. | It is not a CRM; it connects to Salesforce and HubSpot. | Not included; it is another product (Freshsales). | Its great strength: a single CRM for sales, marketing and service. | Customer profile next to the conversation; integrates with HubSpot. | Whatever you build, or connect the CRM you already have. |
| Health of each account | Health score per company, with a generated explanation and alerts. | Not its focus. | Not its focus. | Not its focus; Freddy Insights in Enterprise. | Available in the CRM with properties and dashboards. | Reads the account status from your product to answer. | Whatever you build. |
| Knowledge base | You import PDFs, Word files, websites or MCP servers; map of linked articles; publishing to the portal. | Yes, with an AI-powered knowledge base from Suite Team. | Public, private and multilingual help center depending on the plan. | In all plans; versioning from Pro and approval workflow in Enterprise. | Yes, and an agent that turns tickets into articles. | Help center; reads Notion and Google Drive and writes the missing articles. | Whatever you build on top of your documentation. |
| Integrations | |||||||
| Catalog138 | Small and honest: Gmail, Outlook, Zendesk, Intercom, Stripe, Odoo, webhooks, external databases and any MCP server. | More than 1,000 apps in its marketplace. | More than 450 apps and integrations. | More than 1,000 integrations in its marketplace. | Broad marketplace; the figure could not be read. | About 15: Slack, Zendesk, Intercom, HubSpot, Jira, Notion, Shopify, Zapier… | Anything with an API or MCP server; each one is work on your side. |
| Coexists with your helpdesk | Imports tickets from Zendesk and Intercom so you can start without losing history. | It is the helpdesk. | Fin works on top of Zendesk, Salesforce or HubSpot without migrating ($0.99 per resolution, via sales). | It is the helpdesk. | It is the helpdesk inside its platform. | Coexists with Zendesk or Intercom without migrating. | Connects to whatever you have. |
| MCP | Eleva uses MCP servers (Notion, Stripe, Microsoft…). Own MCP server: coming. | Not checked. | Own MCP server and MCP for giving Fin tools. | MCP actions included in all plans. | Not checked. | They state they have their own MCP server. | Native: it is where MCP was born. |
| Data and security | |||||||
| Where data is stored2479 | Servers in the EU. | Option of EEA, US, Australia or Japan; depending on the plan, a paid add-on. | US, EU or Australia, as you choose. | You choose a data center at the start; which regions are available could not be verified. | Frankfurt data center for new accounts since 2021. | EU (AWS Ireland), no transfers outside the EEA planned. | Wherever you deploy it. |
| Where the AI is processed149 | At Ollama Inc. (US). No DPA signed yet; we are resolving it before we start charging. | OpenAI with zero retention or Azure, AWS or Google clouds; QA and advanced agents can run in Europe. | Unnamed external providers, with a contractual ban on training and retention. | Not published on the pages we read. | Not checked. | Amazon Bedrock in EU regions. | At the model provider; some let you set the region. |
| Certifications2479 | None yet. | SOC 2 Type II, ISO 27001, 27017, 27018 and 27701, among others. | SOC 2 Type II, ISO 27001, 27018, 27701 and 42001; AIUC-1 for Fin. | Its security page points to its Trust Center; we did not check them. | SOC 2 Type 2 and SOC 3. | No SOC 2 of their own; they rely on AWS's. | Yours, plus the model provider's. |
| Permissions | Roles and per-action permissions; the AI works with the permissions of the person using it. | Roles and permissions; detail per plan not checked. | Roles and permissions; detail per plan not checked. | Roles and permissions; detail per plan not checked. | Per-user permissions; SAML SSO only in Enterprise. | Not published. | Whatever you build. |
| Getting started | |||||||
| Time until you can use it | Days: you connect your email, import tickets and documents, and start. | Not published; it depends on size. There are implementation partners. | Fin on top of another helpdesk: "in under an hour", according to them. | Free AI onboarding and guided tutorials. | Not published; it offers an onboarding service. | Minutes to get started, according to them. | A prototype in days; something for the whole team, weeks or months. |
| Maturity | Launching: a new product, with no customers in production yet. | More than 80,000 companies. | More than 25,000 companies; more than 1,400 people. | Between 50,000 and 74,000 companies, depending on the page. | More than 300,000 customers. | Early product. | Your team's. |
| Language and support | App available in Spanish, English, Catalan, German, French and Portuguese. We answer in Spanish or English. | Interface in 28 languages; website and help in Spanish. | Website in English only; Spanish interface not checked. | Website in Spanish; product interface not checked. | Website and phone support in Spanish. | Website in Spanish and English. | The one you program. |
Public data checked on Oct 16, 2026 on each company's website. Zendesk, Intercom, Fin, Freshdesk, Freddy, HubSpot, Breeze, Intake, Claude and Claude Code are trademarks of their owners; Elevatia has no relationship with them. "Not checked" means we did not find it on their public website, not that it does not exist.
We start with when we are not the answer
If you recognize yourself in one of these cards, that option will probably serve you better than we do. Combining them is the normal case anyway.
Zendesk
High multichannel volume, voice and contact center, workforce management, regulated sectors that require certifications in the contract.
Intercom
Lots of repeated questions, a goal of automating as much as possible and willingness to pay per resolution. Or adding AI to your Zendesk without migrating.
Freshdesk
High email volume, often B2C or ecommerce, that wants a complete helpdesk now, a cheap start and AI added by usage.
HubSpot
You already live in HubSpot for sales or marketing: having service on the same CRM is worth a lot.
Intake
A small SaaS or ecommerce, under 2,000 conversations per month, and you want the AI to resolve on its own by performing actions in your system.
Build it with Claude Code
Your own technical team, a very specific workflow, or you want support AI to be part of your product.
Elevatia
B2B support by account, with SLA, where the AI drafts with the context of each customer and a person decides. Teams of 5 to 80 people who do not want to build or maintain anything.
How this page is made
- List prices published on each company's website on Oct 16, 2026, excluding VAT, with annual billing when there is an option. Offers, negotiated discounts and startup programs are not included.
- In the calculator, each platform uses its cheapest plan that already includes an AI agent and no add-ons. If a pricier plan includes something you need, the real cost will be higher.
- Dollar prices converted to euros at the ECB reference rate of Oct 16, 2026: €1 = $1.1253.
- Tokens per conversation (14,000 input and 1,400 output) are an estimate of about four model calls per conversation. You can change them in the calculator; we will publish the real measurement.
- Build and maintenance hours are our own estimates, for an experienced person using Claude Code. They are editable.
- Elevatia's price is the one planned for after the launch and may change. Today the product is free.
- We are an interested party. That is why every data point carries its source: check it, and if something is wrong, write to us.
Sources
- Zendesk — pricing — https://www.zendesk.es/pricing/ (checked on Oct 16, 2026)
- Zendesk — trust center — https://www.zendesk.es/trust-center/ (checked on Oct 16, 2026)
- Intercom — pricing — https://www.intercom.com/pricing (checked on Oct 16, 2026)
- Intercom — security — https://www.intercom.com/security (checked on Oct 16, 2026)
- Freshdesk — pricing — https://freshdesk.com/pricing (checked on Oct 16, 2026)
- HubSpot — Product & Services Catalog — https://legal.hubspot.com/hubspot-product-and-services-catalog (checked on Oct 16, 2026)
- HubSpot — security — https://hubspot.com/security (checked on Oct 16, 2026)
- Intake — precios — https://www.intake.es/es/precios (checked on Oct 16, 2026)
- Intake — seguridad — https://intake.es/seguridad (checked on Oct 16, 2026)
- Anthropic — API pricing — https://platform.claude.com/docs/en/about-claude/pricing (checked on Oct 16, 2026)
- Anthropic — Claude plans — https://claude.com/pricing (checked on Oct 16, 2026)
- OpenAI — API pricing — https://developers.openai.com/api/docs/pricing (checked on Oct 16, 2026)
- Google — Gemini API pricing — https://ai.google.dev/gemini-api/docs/pricing (checked on Oct 16, 2026)
- Ollama — Cloud pricing — https://ollama.com/pricing (checked on Oct 16, 2026)
- ECB — euro reference rate EUR/USD — https://www.ecb.europa.eu/stats/eurofxref/eurofxref-daily.xml (checked on Oct 16, 2026)
What people ask us when they see this page
Why does Elevatia use open models and not Claude or GPT?
For cost and control: to draft a support reply with the account context, a well-supported open model performs well and lets us include all the AI in the price. For hard reasoning, frontier models are better, and the platform already supports connecting them.
If a small Anthropic model costs the same as yours, what am I paying for?
Everything that is not the model: channels, tickets, SLA, accounts, knowledge, permissions, human approval, testing, hosting and maintenance. It is what the "Build it yourself" section breaks down in hours.
Can I use Claude Code with Elevatia instead of choosing one?
That is the idea. We are working on MCP access so your agents can work with your Elevatia data, with your permissions and with approval for anything sensitive. It is built and not yet open.
Why is Elevatia more expensive than Intake at some volumes?
Because we charge more per conversation. In exchange we include SLA, customer portal, CRM and human approval, and we have no volume cap. If you do not need that, Intake may be a better option for you.
Do the prices include VAT?
No. All are excluding VAT, as each company publishes them. In some cases their website does not make clear whether it is included; we note it in the source.
Try it free for as long as the launch lasts
No card. If in the end it is not for you, we will tell you which of these options fits you best.