Pricing last verified: April 2026. Plans and pricing may change -- check the vendor site for current details.
Pricing Overview
Flyte is fully open-source under the Apache 2.0 license with over 80 million downloads. There is no license fee to download, deploy, or run Flyte in production. The project is maintained by the Linux Foundation and will remain free to use indefinitely.
For teams that want managed infrastructure, Union.ai offers a commercial managed platform built on Flyte. Union.ai provides two paid tiers: the Team plan at $950/month (which includes a $950 usage credit) and an Enterprise plan with custom pricing, volume discounts, and dedicated support. This dual model gives organizations a clear path from self-hosted open-source to fully managed without changing their workflow code.
Plan Comparison
| Plan | Price | Compute Included | Key Features | Limits |
|---|---|---|---|---|
| Flyte (self-hosted) | Free / Open Source | None (bring your own infra) | Full orchestrator: workflows, tasks, type-safe data passing, versioning, Kubernetes-native execution, web console | Unlimited users, unlimited workflows |
| Union.ai Team | $950/month | $950 usage credit included | Managed Flyte, GPU scheduling, multi-tenant isolation, 30-day data retention, usage dashboard | Up to 1,000 concurrent actions, 30-day retention |
| Union.ai Enterprise | Custom pricing | Custom allocation | Everything in Team plus multi-cluster, 1-year data retention, volume discounts, dedicated support, SSO, RBAC | Custom limits, SLA-backed |
The Team plan effectively starts at $0 net cost if your compute usage stays within the $950 credit. Once you exceed the credit, compute is billed at the rates below.
Hidden Costs and Considerations
Union.ai charges for compute resources beyond the included credit at granular per-unit rates:
- GPU compute: T4g at $0.15/hr, H200 at $1.58/hr, B200 at $2.85/hr. GPU-heavy ML training pipelines can accumulate significant costs quickly -- a single H200 running 8 hours/day for a month costs roughly $379.
- CPU and memory: $0.0417/vCPU/hr and $0.0051/GB/hr. A workflow using 8 vCPUs and 32 GB memory for 8 hours costs approximately $3.98.
- Self-hosting Kubernetes costs: Running Flyte yourself requires a Kubernetes cluster, an object store (S3/GCS), and a PostgreSQL metadata database. A minimal production setup on AWS costs $300-$600/month in infrastructure before factoring in engineering time.
- Operational overhead for self-hosted: Flyte upgrades, Kubernetes node management, and monitoring require dedicated platform engineering. Budget at least 15-20% of one engineer's time for Flyte operations.
- Data retention limits: The Team plan retains execution data for only 30 days. Teams needing longer audit trails must upgrade to Enterprise or build external logging.
Cost Estimates by Team Size
Solo ML engineer or small research team (1-3 people): Self-hosted Flyte on a small Kubernetes cluster costs $300-$600/month in infrastructure. Union.ai Team at $950/month with the included credit covers light GPU workloads. For teams running fewer than 6 hours/day of GPU tasks on T4g instances, the credit covers the full compute bill.
Mid-size ML team (5-10 engineers): Union.ai Team at $950/month plus an estimated $500-$2,000/month in compute overages depending on GPU usage. A team running 4 H200 GPUs for 4 hours daily would spend roughly $950 + $759 = $1,709/month. Self-hosting at this scale requires $500-$1,200/month in Kubernetes infrastructure plus dedicated platform engineering time.
Enterprise ML platform (20+ engineers): Union.ai Enterprise with custom pricing and volume discounts is the practical choice. At this scale, self-hosting Flyte demands a full-time platform team and $2,000-$5,000/month in infrastructure. Enterprise features like multi-cluster deployment, 1-year data retention, and SSO become necessary rather than optional.
How Flyte Pricing Compares
Flyte competes primarily with other open-source ML workflow orchestrators. MLflow and Kubeflow are both free and open source, with no first-party managed commercial offering at a comparable price point. Metaflow, also Apache 2.0 licensed, is free to self-host.
Weights & Biases takes a different approach: a free tier for individuals, a Pro plan at $60/month per user, and Enterprise pricing on request. For a 10-person team, W&B Pro costs $600/month -- less than Union.ai Team but without the workflow orchestration capabilities that Flyte provides.
Amazon SageMaker uses pure usage-based pricing with no fixed monthly fee, but costs scale rapidly with compute and storage. SageMaker and Flyte serve overlapping but distinct use cases: SageMaker is a full ML platform, while Flyte focuses on workflow orchestration and reproducibility.
Union.ai's $950/month Team plan is positioned above the free open-source alternatives but below full-platform commercial tools. The included $950 compute credit makes the effective cost $0 for light workloads, which is a compelling entry point for teams ready to move beyond self-hosted Flyte without committing to enterprise contracts.