Model Your Agent Platform Credit Burn Before the Free Allowance Ends
The complimentary Agent Platform (formerly Opal) allowance ends December 31, 2026. Until then, each instance can use up to 200 credits a month at no cost, and Optimizely reserves the right to change that sooner.
200 credits is about one and a half runs of Page Builder, which averages 130 credits per invocation.
If your team has been working inside that allowance, or quietly past it, January is when the meter starts to matter. The good news is that the credit model is simple enough to estimate on a napkin. Here's how it works, a fast way to model your monthly burn, and what to ask your account team before the new year.
How credits work
Every agent is assigned a complexity band that sets its average cost per invocation:
| Band | Avg credits per invocation |
|---|---|
| Basic | 2 |
| XS | 10 |
| S | 30 |
| M | 70 |
| L | 130 |
| XL | 200 |
A few published per-agent averages that can help inform content and marketing teams:
- Page Builder: 130
- Content Analysis: 120
- SEO Metadata Implementation: 120–200
- SEO Metadata Optimization: 50–100
- Image Alt Text Generation: 40
- Content Summary, Email creation, Content Translation, UTM Creation: 2
These are averages. Actual cost varies with input size, context and model usage.
Three rules govern the math:
- Workflows cost the sum of their agents. There's no overhead for the workflow itself, and no discount either. A workflow that chains four agents costs four agents.
- Credits are pooled. Every eligible Optimizely product draws from the same pool.
- There are no per-user limits. Optimizely doesn't support them. The only control is access: users without Agent Platform access (the docs currently call it Mark access) don't consume credits.
Back-of-envelope math
Assume about 21 business days a month.
| Scenario | Math | Credits / month |
|---|---|---|
| Page Builder once per business day | 130 × 21 | 2,730 |
| SEO Metadata Implementation on 20 pages | (120–200) × 20 | 2,400–4,000 |
| Workflow: Content Analysis + SEO Metadata Optimization + Image Alt Text, run 10× | (120 + 50–100 + 40) × 10 | 2,100–2,600 |
| 10 people, 5 basic-agent runs each per business day | 10 × 5 × 2 × 21 | 2,100 |
Each scenario on its own is 10 to 20 times the free allowance. A team doing all four lands around 9,300–11,400 credits a month, or roughly 47 to 57 times the allowance.
Two lessons fall out of this: First, model per job, not per agent. Teams think "we use Page Builder," but the bill comes from the workflow that calls Page Builder plus three other agents. Second, cheap agents aren't free at scale. Two credits a run disappears in the noise until ten people use it all day.
One more thing to watch is scheduled and triggered workflows. Credits are attributed to whoever schedules the workflow or sets up its webhook or email trigger, which means a workflow can burn credits every day without anyone opening chat.
Do this before January 1
- Pull your real numbers. Optimizely Reporting's Opal Agent Usage dashboard shows credits per agent and Avg Credits / Exec. Compare those against the published averages, because your inputs aren't average. Reporting figures are informational. For the billing view, use Usage & Billing in the Opti ID Admin Center, which requires the Super Admin or Billing Admin role.
- Turn on credit usage notifications. In the Admin Center, go to Settings > Notifications. You get a monthly notice, plus extra notices once usage passes 100% of your allotment.
- Decide who has access. With no per-user caps, your access list is your budget control. Audit it like a license list. Admins can also turn off generative AI across Optimizely applications if a product shouldn't use it.
- Find your expensive agents and workflows. Multiply Avg Credits / Exec by execution count and sort. Look hard at anything scheduled.
Questions for your Optimizely account team
- What's the price per credit, and is it locked for the term? Pricing is quote-based and isn't published.
- Can an agent's band assignment change mid-term? The credit value model was already adjusted once, effective March 1, 2026. Band assignment drives cost as much as price per credit does.
- What happens operationally when we hit our limit? The docs say admins are notified and that usage beyond the allotment is still tracked. They don't say whether anything stops.
- Do unused credits roll over or expire? The billing dashboard shows an expiry date for each allocation. The docs don't say whether anything carries forward.
- How are overages billed? Ask about the rate, the timing of any true-up, and whether there's a cap.
- Do chat-only exchanges consume credits? If your team uses chat heavily without invoking agents, ask. The current credits page doesn't say clearly.
Cost governance is access governance
Deciding who can run which agents is the same decision as deciding what you'll spend. Make both decisions in one meeting, with the usage report open, before the allowance goes away.
At WOW, we're helping clients build this model from their own Reporting data, so the first conversation with their account team starts from real numbers instead of averages.
Sources