Overview
Pricing model
Stac runs on dedicated GPU capacity, not on a per-token meter — what a fixed monthly price covers, and what happens when you outgrow a tier.
Stac runs on dedicated GPU capacity, not on a per-token meter. You pick a plan based on the model and features your product needs, and that capacity is yours — reserved, provisioned, and billed monthly, in advance.
What the price covers
Every plan includes:
- A GPU capacity slot for your model, always on
- Unlimited requests and unlimited tokens — no per-token charges, no usage-based surprises
- A dashboard with usage logs and access to your keys
The model field in your request is ignored — Stac always routes to the
model your plan is running. You never need to touch client code when we
upgrade the underlying model for your tier.
What varies
Plans stack on top of each other: each tier includes everything in the one below it, plus a bigger model, more document search capacity, and more control over how the AI behaves.
Tier
| Plan | Price | Includes |
|---|---|---|
| Go | $25/month | AI model ready to use · Custom instructions for your product · Document search (RAG) |
| Pro — Most popular | $65/month | Everything in Go · Larger, more capable AI model · Expanded document search (RAG) · Full control over AI response behavior |
| Max | $180/month | Everything in Pro · Fine-tuning on your own data |
| Enterprise | Custom (unlimited repos and seats) | Custom-sized capacity for your volume · Unlimited fine-tuning with your own data · LGPD compliance via contract · Full audit logs · Access controls by user/team · Priority support with SLA · Dedicated account manager |
Document search (RAG) limits
| Plan | Files | Max size per file |
|---|---|---|
| Go | Up to 5 | 10MB |
| Pro | Up to 20 | 10MB |
| Max | Up to 50 | 10MB |
| Enterprise | Custom | 10MB |
Requests are never cut off
Two things Stac does specifically so you never get a truncated or rejected response, regardless of your plan:
- No hard token ceiling. Stac enforces a floor under the hood on how many tokens a response is allowed to generate, so a low or missing value in your request never truncates the model mid-reasoning before it reaches its answer.
- Elastic concurrency, not a fixed reject. There's no per-key request cap. If you burst above your plan's usual throughput, requests queue and complete — they don't fail with a hard error.
Changing tiers
- Upgrading takes effect immediately, with a prorated charge for the rest
of the billing period — the
modelfield you send stays the same, Stac just points it at something more capable. - Downgrading takes effect at the start of your next billing period; you keep what you already paid for until then.
Enterprise
Need custom-sized capacity, unlimited fine-tuning, audit logs, or a dedicated account manager with an SLA? Enterprise plans are negotiated directly — talk to us.