$GPU Rental Prices.com

Vast.ai GPU pricing

Visit Vast.ai →

Vast.ai is a GPU marketplace where hosts set prices and buyers rent GPUs through the console, CLI or API. It fits cost-sensitive builders who can compare offers, reliability scores and bandwidth terms instead of buying from a fixed-price cloud catalog. Billing is per second, with on-demand, interruptible and reserved options plus separate storage and bandwidth charges.

Live prices for Vast.ai are not in the tracker yet: the profile below is verified editorial. Tracking is being added.

Pricing model

Vast.ai uses marketplace pricing: hosts set rates and the market changes in real time. The pricing docs say total cost combines GPU compute, storage and bandwidth, and usage is billed by the second. Serverless has no separate pricing tier in the docs checked; users pay the underlying instance costs. Billing requires prepaid credits for rentals, and the billing page shows itemized charges for GPU usage, reserved disk storage and upload or download usage.

Capacity tiers

Vast.ai documents three instance types. On-demand is high priority, fixed-price access with guaranteed resources for the rental period. Reserved is high priority with prepaid discounts and is created by converting an on-demand instance. Interruptible is low priority, uses a bidding system and may be paused if outbid or if on-demand capacity is requested. The public pricing page also describes on-demand, interruptible and reserved as the main pricing tiers.

GPU lineup

Vast.ai publishes broad GPU coverage rather than a fixed datacenter-only catalog. Its GPU Cloud page lists Blackwell RTX 5090, Hopper H100 NVL, H100 SXM, H200 and H100 PCIe, Ada Lovelace RTX 4090, L-series and RTX Ada options, Ampere RTX 3090, RTX A-series and A100, plus older Volta, Tesla, Turing, Pascal and Maxwell families. The pricing page also says the platform spans many GPU types across data centers.

Storage, egress and hidden costs

Vast.ai's billing docs are explicit that active rental, storage and bandwidth are separate charges. Active rental is charged while an instance is active, storage is charged for every second an instance exists and is online, and stopping an instance does not avoid storage costs. Bandwidth is charged for every byte sent or received, with upload and download rates varying by machine. Buyers should inspect the offer's price breakdown before renting.

Reliability and support posture

Vast.ai is a marketplace, so reliability depends on the selected host, tier and offer. The pricing docs say higher reliability scores usually correlate with higher prices, and the instance docs recommend checking maximum duration and host reliability for long-running jobs. Vast.ai's GPU Cloud page states it offers 24/7 human support, premium support tiers, onboarding, architectural consulting and guaranteed response times for higher tiers. Secure Cloud pages state SOC 2 Type II compliance for that environment.

Free credits and referral programs

Vast.ai says there is no current free trial on its launch page. The startup page documents a credit matching program where approved companies receive credits matched against verified compute spend, with no upfront giveaways and no free trials. Vast also documents a referral program: new client accounts that buy credits through a referral link can generate lifetime referral credits, with cash-out rules and dedicated-account requirements. See Vast startup program and Vast referrals.

Who should pick Vast.ai

Pick Vast.ai if you are comfortable shopping a live marketplace and want a wide spread of GPUs, host types and reliability profiles. It is a good fit for checkpointed training, experiments, batch jobs and teams that can evaluate bandwidth and storage costs before launch. Use on-demand or reserved for work that cannot pause, and interruptible only when the job can resume cleanly. For production or sensitive workloads, filter for higher-trust capacity and confirm support terms.

FAQ

Is Vast.ai a fixed-price cloud?

No. Vast.ai is a marketplace where hosts set prices and buyers compare real-time offers.

Does Vast.ai bill by the second?

Yes. Vast.ai docs say GPU usage is billed by the second for actual usage.

Can Vast.ai interrupt instances?

On-demand and reserved are high priority. Interruptible instances can be paused if outbid or if on-demand capacity is requested.

Does Vast.ai have a free trial?

No public free trial is documented. Vast.ai documents startup credit matching and a referral program instead.

Profile verified 2026-07-06. Sources: vast.ai · vast.ai · vast.ai · docs.vast.ai · docs.vast.ai · docs.vast.ai · docs.vast.ai · docs.vast.ai · vast.ai · vast.ai