Every plan starts free for 30 days.
Sign up, run real workloads, pay nothing until day 31.
For teams building AI from experiment to production.
For organizations building mission-critical AI, ML, and agents.
“Union.ai has been critical at the time that we needed to significantly scale up… Union.ai’s wealth of expertise has enabled us to move fast and deliver at scale.”
You pay for what each action consumes: the action itself, plus the vCPU, memory and GPU it allocates, metered to the second. Every ladder is measured on the volume you use in a calendar month and starts again from the first band on the first of the next. Nothing carries over.
One durable unit of work. A task, a trace, a condition or a cache hit each count as one, however long it runs and however many times it retries.
Allocated vCPU on every container running your actions, read from the task’s resource request and billed to the second.
Allocated memory, same basis as vCPU. Bands are six times the vCPU bands, so a normally shaped workload crosses both meters together.
Bands are marginal. A month of 2.5M actions is priced as three slices, not one rate:
Every accelerator belongs to one of three classes, set by what the silicon costs on the open market. The class rate is the same on AWS, GCP, Azure and neoclouds.
| Class | 0 – 1k hrs / mo | 1k – 10k | 10k + |
|---|---|---|---|
| Class 1 Accelerators under $1.50 / hr on the open market T4T4gL4Inferentia2TPU v5eTrainium1 | $0.12 | $0.10 | $0.08 |
| Class 2 $1.50 – $5.50 / hr A10GL40SV100A100RTX PRO 6000Trainium2TPU v6eTPU v5p | $0.38 | $0.32 | $0.26 |
| Class 3 Over $5.50 / hr H100H200B200B300 | $1.20 | $1.00 | $0.80 |
| Reserved or owned GPUs Capacity reservations, committed-use contracts, or your own silicon | Talk with us | ||
Enterprise volume discounts. The ladders above are the published Team rates. Enterprise plans are priced on commit, with significant discounts across actions, compute and GPU.
Explore with usManaged deployment (BYOC), optional. Union hosts the control plane and runs the data plane inside your cloud for you. Self-managed, where you run the infrastructure, carries no fee. The fee sits outside the plan minimum.
Compare deployment options| Feature | Team | Enterprise |
|---|---|---|
| Platform | ||
| Volume discounts | Published ladders, applied automaticallyBands reset monthly | Published ladders plus commit discounts |
| Monthly usage credit | $950 | Custom |
| Data retention | 30 days | 1 year |
| Price / action | from $0.0030to $0.0003 at 10M+ | Custom on commit |
| vCPU · memory · GPU | from $0.0288 · from $0.0040 · from $0.12 | Custom on commit |
| Managed deployment (BYOC) | $2,500 / env / mo, optional | $2,500 / env / mo, optional |
| Self-hosted control plane | — | On-prem, hybrid, air-gapped |
| Durability & debugging | ||
| Advanced debugging, automatic retries | ||
| Output caching | ||
| Task-level overrides | ||
| Observability | ||
| Live remote debugging | — | |
| Scalability | ||
| Actions / run (fanout) | 25,000 | 50,000+ |
| Concurrent actions | 1,000 | Custom |
| Clusters | 1 | 3+ |
| Security & support | ||
| Zero Trust security | ||
| SSO | Standard (OIDC) | Custom (OIDC, SAML/p) |
| Role-based access control | Standard | Fine-grained, custom |
| Support | 1 business day Sev 1 | Custom white-glove support |
An action is one durable unit of work: an invocation of flyte.task, flyte.trace, or flyte.new_condition running end to end to completion. A cache hit also counts as one action.
Each meter has a published ladder. The first units you use in a month are priced at the first band, the next at the second, and so on. The lower rate applies only to the units above the threshold, so crossing a band never makes the units below it more expensive.
On the first of every month. Bands are measured on the volume consumed in a calendar month and start again from the first band the next month. Nothing carries over, in either direction.
Every self-service signup starts with a 30-day free trial, running from the day you sign up. On day 31 the Team plan begins: a $950 monthly minimum, fully credited against your usage, billed through AWS Marketplace.
Your monthly plan fee is issued back to you as usage credits. In practice, this means your monthly plan cost becomes your minimum monthly spend, and you can use that same amount in usage at no additional charge. Any usage beyond that amount is billed separately.
No. An action is counted regardless of time, whether it finishes in two seconds or runs for a month. Compute is metered separately, to the second, on the resources the action allocates.
No. Actions are counted only once the task, trace, or condition succeeds, whether that happens on the first try or after five retries.
No. Union’s built-in durability layer is inherent to all actions and doesn’t count toward action usage. Whether Union checks on a task once or a hundred thousand times, that’s still one action.
We report the allocated resources (CPU, memory, and GPU accelerator) from each container running the actions within your workflows and apply usage-based pricing down to the second.
With managed deployment, Union runs the data plane inside your cloud account as well as hosting the control plane: provisioning, upgrades, and operations. It is $2,500 per environment per month and is optional. If you would rather run the infrastructure yourself, self-managed deployment carries no fee.
Fanout is the total number of actions a run creates, while concurrency is how many of those actions are running at the same time.
Yes we do! Talk to our Startup Team to learn more.
Yes we do! Talk to our Public Sector Team to learn more.
Yes, this is available as part of a Union Enterprise plan.
Yes, this is one of the deployment options available as part of a Union Enterprise plan. Self-hosted deployment supports on-prem, hybrid, and air-gapped configurations.
Zero Trust means your data never transits Union.ai’s infrastructure. Workflow inputs, outputs, logs, and code bundles are served directly from inside your own cloud through a secure tunnel.
No. You deploy the Union operator into a Kubernetes cluster you manage, which securely communicates with the Union control plane to poll for work.
Yes, Union.ai supports bring-your-own-cloud (BYOC) deployments. You can run it in your own AWS, GCP, Azure, or neo-cloud environment.