Vellum AI Pricing in 2026 by Seats, Usage and Tokens
Oct 29, 202613 min read
Vellum AI pricing: current plans, prices and limits
Vellum AI pricing summary and date checked
No single plan fee tells you what Vellum costs, much as teams comparing it with workflow tools like Activepieces find that a headline number rarely survives contact with real usage. The price depends on your seat count, usage limits and model tokens.
Vellum is a platform for building, testing and running LLM applications and agents, so a bill is really a sum of several meters. I wrote this page on 29 October 2026 and left out any dollar figure I couldn't confirm.
Plan table: price, included features and limits
Ignore the plan names; they are the least useful part of the pricing page. To see which lines scale with your workload, read what each plan meters in the table below.
Ignore the plan names; they are the least useful part of the pricing page.
A line that scales with your team or traffic matters more than the headline fee.
| Meter | What is counted | Effect on your bill |
|---|---|---|
| Seats | Each person who builds or edits workflows | Cost rises with headcount, so a contractor you add for one project is billed like a full-time engineer |
| Usage limits | Runs or executions allowed per period | A busy workflow can hit the cap mid-month and push you to a higher plan |
| Model tokens | Text the underlying model reads and writes | Billed separately from the platform, so a cheap plan can sit on top of an expensive model bill |
| Retention and gated features | How long logs are kept, plus SSO, role-based access control (RBAC) and private-network (VPC) deployment | Security requirements can force you into a higher tier regardless of volume |
Price each row against your own numbers before you compare tiers. A team of four with light traffic and a team of twenty with heavy traffic are buying different products under the same plan name.
Which prices are published and which need a sales call
Expect Vellum to publish its entry pricing. Anything tied to security controls or large-scale deployment is most likely quote-only, so treat it as "contact sales" until a rep gives you a number in writing.
A security review will ask for the gated features. Activepieces charges 1 credit per flow run however many steps the run contains, so a ten-step flow costs the same as the two-step workaround it replaced, where Zapier bills per task.
The run is the meter, so splitting work into smaller steps never raises the price.
If you are running this arithmetic for your own team, see what the same workload costs on Activepieces.
What Vellum AI plans cost at realistic usage
Double the usage and it is model tokens that double, and that's why Vellum's platform fee shrinks from about two-thirds of a solo bill to roughly 6% of a compliance team's, so solo users face much steeper costs.
The runs and tokens below are our assumptions, priced at a blended $5 per million tokens, with included credits offsetting token spend.
| Scenario | Plan | Platform fee | Assumed runs and tokens (ours) | Provider token cost | Total | Platform share |
|---|---|---|---|---|---|---|
| Solo prototype | Mighty | $30 | 2,000 runs, 8M tokens | $15 after credits | $45 | 67% |
| Five-person team | Super | $100 | 20,000 runs, 100M tokens | $455 after credits | $555 | 18% |
| Larger compliance team | Ultra | $200 | 100,000 runs, 600M tokens | $2,885 after credits | $3,085 | 6% |
viz type=groupedbars title="Vellum paid plans: price against included model credits" unit="$ per month" source="AI Provider Index - https://aiproviderindex.com/platform/vellum" cols="Plan price,Included usage credits"00, so token spend lands on top." Mighty: 30, 25 Super: 100, 45 Ultra: 200, 115
Price your own run count against these rows, because the fee barely moves while the token line does.
Vellum AI cost for a solo builder
$30 per month gets you Mighty, with $25 in model credits, per the AI Provider Index. Our 8M tokens would cost $40, so the credits absorb most of it and the plan fee is the main expense.
Vellum AI cost for a five-person team
For $100 per month, Super brings $45 in credits, per the same index. Against $500 of assumed token spend, the credits cover under a tenth, so the fee is a minor line and the provider invoice sets the budget.
Vellum AI cost for larger secure teams
At $200 per month with $115 in credits per the index, Ultra includes a monthly usage allowance, but most of your monthly bill still varies by usage. We assume this is the tier a compliance team would need.
At $3,000 of assumed token spend, the credits cover under 4%, so doubling usage roughly doubles the bill while the platform fee stays flat.
Hidden costs and limits Vellum AI buyers hit
Beyond the platform fee, five costs can move a Vellum bill, and only some of them have a published number. These are the ones to check against your own workload:
- Model tokens billed outside the plan: the credit math above shows it, and it grows with usage.
- Seat caps: unstated, so ask what the plan allows before adding teammates.
- Usage and retention limits at cap: unstated, so you cannot yet say what happens when you hit them.
- SSO, RBAC and VPC gated to higher tiers: unstated, so assume a security review adds cost.
- Sales-led pricing: unstated for the top tier, so the quote is the price.
Each is worth a closer look.
Model tokens are a separate bill
You pay twice here. Vellum charges for the platform, and the model provider charges for tokens. Because the credits cover under 4% of the assumed $3,000 spend, your token volume sets the total.
Compute add-ons and usage limits
Look at the compute add-ons and the steps get steep. AI Provider Index lists monthly prices for larger sizes. Medium is $35 a month, so one modest upgrade barely registers.
Large is $60, nearly double Medium. XL is $125, more than triple Medium, so a workload that outgrows the small sizes pays a steep step for each jump. Multiply by the number of environments that need it.
Vellum's usage and retention limits at cap are unstated, so confirm them before relying on logs or history.
Features gated behind higher tiers
Before anything else, security teams ask for SSO, role-based access control (RBAC) and private-network deployment (VPC). Which tier includes each is unstated here, so a buyer who needs them may be priced on a higher plan than the one they compared.
Pricing that only appears after a sales call
Where a tier shows no price, the number comes from a quote. You can't model its cost in advance, so give the vendor your seat count, run volume and token estimate, and ask for a written figure.
Worth checking against a plan that does not meter every step: one credit covers a whole run on Activepieces.
Vellum pricing versus LangSmith, Langfuse and Dify
Asking for a written quote matters most with Vellum because, unlike Langfuse and Dify, the other three let you avoid a vendor bill entirely by hosting the software yourself.
Comparison table on price, metering and self-hosting
Which of your own numbers (people, runs or traces) will drive the bill depends on the unit each vendor meters and where it puts its gates, and the table shows both.
Wording is qualitative on purpose: re-read each vendor's pricing page on the day you buy, since these plans change.
| Free tier | Entry paid plan | Metering unit | Self-hosting | Enterprise gating | |
|---|---|---|---|---|---|
| Vellum | Limited starter access | Per-seat plan | Seats plus workflow runs; model tokens billed separately | Arranged through sales | Larger limits and isolation by quote |
| LangSmith | Single-developer tier | Per-seat plan | Seats plus traces | Enterprise plan only | Self-hosting and custom terms |
| Langfuse | Cloud free tier | Base fee plus usage | Traces and observations | Open source, run it yourself | Some governance features |
| Dify | Cloud sandbox | Per-workspace plan | Message credits and team members | Open source, with license conditions | Multi-workspace and SSO-type features |
Where Vellum costs more and what you get for it
Pay Vellum's premium and you get a managed workflow builder, prompt versioning and evaluation in one product. If your team would otherwise assemble those integrations, the premium buys fewer integration hours.
If a single engineer owns the pipeline, you're paying per seat for tooling you may not open often.
Where an alternative is the cheaper fit
When you only need tracing and can run the open-source version on infrastructure you already pay for, Langfuse is cheaper. Dify is cheaper when non-engineers build simple apps and message credits, not seats, are your limiting unit.
LangSmith is cheaper than Vellum for a small team already tracing LangChain code, because it meters traces rather than workflow runs.
Activepieces for LLM steps inside business workflows
Where LangSmith meters traces, Activepieces meters runs, at 1 credit per run however many steps it has, with AI steps costing extra credits and unlimited flows on every plan, including free. That fits when an LLM call is one step in a business automation.
It doesn't replace Vellum for prompt evaluation.
When the task is automation, not prompt engineering
Choose Activepieces when the model call sits between ordinary steps, across its 746 integrations:
- Email routing: a message arrives and the model classifies it.
- An LLM call: summarize or draft inside the flow.
- A CRM update: the result is written to the customer database.
- Self-hosting: you want the automation engine on your own infrastructure.
Stay on Vellum when the prompt itself is the product:
- Prompt evaluation against test cases.
- Versioned prompt testing, so you can compare revisions.
- Agent deployment.
Activepieces' free plan lists Flows, Agents, Tables, MCPs and API access, and neither prompt evaluation against test cases nor versioned prompt testing appears there, so moving those jobs over means rebuilding the tooling yourself.
Estimating credit usage by model
Pick the model and you pick how fast credits drain, because the credit is the meter.
On a frontier model, free credits cover only 50 of 1,000 steps, leaving 950 to pay for. On a fast model they cover 500 of the 1,000, leaving the other 500 to pay for.
So a workflow that runs a frontier model on every email exhausts the free allowance in a day or two, while the same flow on a fast model lasts ten times longer.
The self-hosted Community Edition is MIT-licensed, so run it yourself and the automation engine moves off the vendor's meter, though you still pay your model provider for tokens and carry the hosting work.
Check the plan prices on its pricing page against your step count before you commit.
Count steps per run, multiply by monthly runs, then decide which model each step needs. Only the steps that truly need a frontier model should use one.
Vellum AI pricing checklist before you buy
Once you know which steps need a frontier model and which can run on a cheaper one, the buying order is fixed: count your workload first, then ask what each count costs.
Work through these in sequence, because each step gives the next one its inputs.
- Estimate monthly runs and the tokens each run consumes.
- Get limits and overage terms in writing.
- Confirm which tier includes each feature.
- Trial with your real workload, not a demo.
- Set spend alerts at the model provider.
- Take the results to Vellum sales.
Estimate your Vellum AI usage and tokens
Pick the unit Vellum meters, whether that's runs, seats, or something else, and count how many of those your team generates in a normal month.
Then, for each step in a run, note the model it calls and roughly how many input and output tokens it uses.
Multiply through. The result is your bill's real shape, and the plan names on the pricing page can't show you that.
Get limits and overage terms in writing
Before you rely on a limit, know what happens when you hit it. Ask whether the platform throttles you, blocks you, or bills you more. Ask whether unused allowance rolls over.
A sales call isn't a contract, so request the answers in an order form or email you can point to later. If you need an exception to a limit, that's the moment to negotiate it.
Questions to ask Vellum sales
Bring your volume estimate and ask:
- Which tier includes every feature on my shortlist, and which are add-ons?
- Are model tokens billed by Vellum or directly by the provider, and at what markup, if any?
- How is a seat defined, and do contractors or read-only reviewers count?
- What does a run cost when it fails and retries?
Their answers set the final figure, and the trial results tell you whether to believe them.
Frequently asked questions
Where the pricing body leaves gaps, these five questions pick them up. Each answer ends with where it comes from, and none comes with a figure.
Does Vellum AI have a free plan?
Unconfirmed here. Even if a free tier exists, ask what it meters, whether that's seats, runs or workflow executions, because the meter decides how quickly you outgrow it and what the next tier costs. Source: unverified.
Does Vellum AI charge for the models it runs?
Plan on paying for model tokens separately from the platform fee. That's the premise of this article, and it means your workload's prompt length and call volume set that line, not the plan name.
Whether Vellum passes provider rates through or lets you bring your own keys is unverified, so ask before you model the cost. Source: unverified.
Can you self-host Vellum AI?
Treat self-hosting as a question for sales, not an assumption. If it exists, it's likely tied to a higher tier and a separate contract, so price it as its own line against the hosted plan. Source: unverified.
Does Vellum AI offer discounts for startups or annual billing?
Ask for both in writing. Annual commitments change what you owe if your usage shrinks, because you pay for capacity you may not use. A startup program, if one exists, is worth asking about during the trial, before you commit. Source: unverified.
How often does Vellum AI change its pricing?
There's no confirmed change history. The practical protection is to screenshot the pricing page the day you run your trial and attach it to your quote.
Then any later difference in seat counts, usage limits or token billing is something you can point to. Source: unverified.
Related reading
References
Running the numbers
See what the same workload costs here.
Free forever plan, and every paid plan self-hosts at no extra cost.
See pricing Talk to sales
