Vast.ai vs Runpod: GPU Prices, Storage and Fees
Vast.ai vs Runpod verdict: Vast.ai is the better pick when you shop each rental for the lowest host ask or need interruptible bids; Runpod is the better pick when you want one listed price per GPU, no bandwidth fees and cheaper list storage. Neither wins on every card. At 10:00 Europe/Paris on November 10, 2026, Vast.ai’s median verified ask beat Runpod’s Community Cloud list price on A100 80GB cards, while Runpod Community was lower on the RTX 4090 and H100 SXM — and Vast.ai’s cheapest H100 SXM ask still undercut both.
Pick Vast.ai if: you compare offers before every job, run checkpointed work that can use interruptible bids, or want to prepay one host for a reserved discount without a sales call.
Pick Runpod if: you move large datasets in and out (no ingress or egress fees), want Secure Cloud data centers at a fixed list price, or keep volumes parked between sessions at $0.20 per GB per month.
- Price model: Vast.ai is a marketplace where hosts set GPU, storage and bandwidth rates; Runpod publishes Community and Secure Cloud list prices (page marked Updated November 4, 2026).
- Hidden meter: Vast.ai bills bandwidth per byte at host rates; Runpod Pods docs state no fees for data transfer.
- Interruptible: Vast.ai sells bid-priced interruptible instances; Runpod’s current Pods pricing docs list on-demand only, with savings plans deprecated.
By the Unfair Advantage editorial team · Updated November 10, 2026. Method: Vast.ai rates come from unauthenticated queries to its documented search offers endpoint at 10:00 Europe/Paris on November 10, 2026 (verified, rentable, unrented, single-GPU, on-demand; a second pass limited to datacenter offers; A100 filtered to 80 GB and RTX 4090 to 24 GB to match Runpod’s cards). Runpod rates come from its public pricing page fetched at 09:57 the same morning (card data for Community Cloud and Secure Cloud; page marked Updated November 4, 2026). Rules come from both platforms’ documentation and Terms fetched the same morning. We did not open an account, rent a GPU or run a benchmark on either platform. Runpod’s live console price after login was not checked. Vast.ai medians are our arithmetic on the offers returned (at most 64 per query). For a full platform verdict on Vast.ai alone, see our live Vast.ai review on cheap GPUs and host trade-offs.
How do Vast.ai and Runpod differ at a glance?
Vast.ai and Runpod both rent GPU containers billed by the second, but Vast.ai is a host marketplace with three variable meters and Runpod is a list-priced cloud with two meters. That single difference drives most of the comparison below.
| Criterion | Vast.ai | Runpod | Edge |
|---|---|---|---|
| GPU price | Set by each host; moves by the minute | Listed per GPU for Community and Secure Cloud | Depends on the card |
| Bandwidth | Per byte, both directions, host-set | No ingress or egress fees | Runpod |
| Storage | Host-set $/GB/month while the instance exists | $0.10 running / $0.20 stopped volume; $0.07 network volume (<1 TB) | Runpod on list price |
| Interruptible | Yes, bid-priced | Not listed in Pods pricing docs | Vast.ai |
| Reserved | Self-serve prepay on one instance; docs “up to 50%” | Savings plans deprecated; reserved capacity via sales | Vast.ai for self-serve |
| Deposit to start | Prepaid credits | At least one hour of credits for the configuration | Tie |
| Spend cap | Not disclosed in docs checked | $80 per hour default, raised over time | Runpod for guardrails |
Should you choose Vast.ai or Runpod?
Choose from your workload’s tolerance for interruption and data movement first, price second.
Vast.ai fits when…
- You will pick a host per job by price, reliability score and bandwidth rate.
- Your training checkpoints, so interruptible bids are usable.
- You want A100 80GB time: Vast.ai’s median verified ask was 19% below Runpod Community on A100 SXM and 27% below on A100 PCIe at 10:00.
- You want to prepay one host for a discount without contacting sales.
Runpod fits when…
- You pull or serve large files: no ingress or egress fees on Pods.
- You need a price you can put in a budget: one list price per GPU and cloud tier.
- You park environments between sessions: volume disk is $0.20 per GB per month when stopped.
- You need an RTX 4090: Runpod Community listed $0.34 per hour against a $0.536 Vast.ai median at 10:00.
Which is cheaper per GPU hour, Vast.ai or Runpod?
Runpod’s Community Cloud list price was lower than Vast.ai’s median verified ask on 8 of the 11 single-GPU models compared on November 10, 2026; Vast.ai was lower on both A100 80GB models and the RTX PRO 6000. Even where Runpod’s list price beat the Vast.ai median, Vast.ai’s cheapest single offer was lower on the H100 SXM, RTX 5090 and RTX 3090.
| GPU (1×) | Vast.ai median verified ask $/h | Vast.ai lowest datacenter ask $/h | Runpod Community $/h | Runpod Secure $/h |
|---|---|---|---|---|
| H100 SXM | $3.073 (lowest $2.138; 6 offers) | $3.877 | $2.69 | $3.99 |
| H100 NVL | $3.376 (lowest $3.376; 1 offer) | none returned | $2.59 | $3.19 |
| H200 | $4.344 (lowest $4.344; 1 offer) | none returned | $3.59 | $5.29 |
| B200 | $12.510 (lowest $12.510; 1 offer) | none returned | $5.98 | $7.99 |
| A100 SXM 80GB | $1.122 (lowest $1.069; 6 offers) | $1.069 | $1.39 | $1.79 |
| A100 PCIe 80GB | $0.869 (lowest $0.869; 3 offers) | none returned | $1.19 | $1.79 |
| L40S | $0.802 (lowest $0.801; 5 offers) | none returned | $0.79 | $1.09 |
| RTX PRO 6000 (96 GB) | $1.558 (lowest $1.402; 12 offers) | $2.470 | $1.69 | $2.49 |
| RTX 5090 | $0.743 (lowest $0.463; 48 offers) | $0.497 | $0.69 | $1.19 |
| RTX 4090 (24 GB) | $0.536 (lowest $0.389; 31 offers) | $1.383 | $0.34 | $0.89 |
| RTX 3090 | $0.256 (lowest $0.116; 27 offers) | $0.369 | $0.22 | $0.50 |
Vast.ai: search offers API, 10:00 Europe/Paris, November 10, 2026. Runpod: pricing page card data, page marked Updated November 4, 2026, fetched 09:57 Europe/Paris the same day.
How to read it: Vast.ai prices are per-offer asks (dph_total, GPU plus the storage the API bundles into the offer price); bandwidth is extra. Runpod prices are card list prices; Community Cloud is labelled “from” on Runpod’s own FAQ, so availability at that price is not guaranteed. Specs differ: Runpod cards fix vCPU and RAM per GPU (for example 6 vCPUs and 41 GB RAM on the RTX 4090), while Vast.ai hosts vary. The RTX PRO 6000 row compares Vast.ai’s workstation edition with Runpod’s unnamed RTX Pro 6000 96 GB card.
Runpod also lists cards outside our Vast.ai query set, such as the B300 ($6.94 Community / $8.99 Secure) and Pro 6000 MIG slices ($0.69 and $1.29 Secure); MIG slices are Secure Cloud only.
How do storage and bandwidth costs compare?
Runpod is cheaper and simpler on storage and bandwidth list prices; Vast.ai can match it only on hosts that price those meters low. Runpod Pods docs state “no fees for ingress/egress,” while Vast.ai charges “for every byte sent or received” at each host’s rate.
| Meter | Vast.ai (10:00, medians) | Runpod (list) | Note |
|---|---|---|---|
| Running disk | $0.20–$0.31/GB/month on A100, L40S, H100 SXM and RTX offers | $0.10/GB/month (container or volume disk) | Vast.ai datacenter H100 SXM median $0.672 |
| Stopped disk | Still billed; docs say often higher when stopped | $0.20/GB/month volume; container disk not charged (erased) | Both bill parked data |
| Shared / network storage | Volumes priced by host | $0.07/GB/month under 1 TB, $0.05 above | Runpod network volumes are portable |
| Egress | $5.33/TB (RTX 4090) to $36/TB (datacenter H100 SXM); one H100 NVL offer $102.40/TB | $0 | Runpod for data-heavy jobs |
Both platforms release the GPU when you stop. Runpod warns that a resumed Pod may get zero GPUs if capacity changed; Vast.ai says a stopped reserved instance’s GPU “may be rented by another user.” Plan restarts accordingly on either side.
What would a month cost on each platform?
A month on an always-on RTX 4090 cost roughly $255 on Runpod Community list prices versus about $418 at Vast.ai’s median ask on November 10, 2026, once storage and egress are included. On H100 SXM time the order flips at Vast.ai’s lowest ask.
| Scenario | Vast.ai | Runpod Community | Runpod Secure |
|---|---|---|---|
| RTX 4090, 720 h, 100 GB disk, 1 TB egress | ≈ $418 (0.536 × 720 + 100 × 0.267 + 5.33) | ≈ $255 (0.34 × 720 + 100 × 0.10) | ≈ $651 (0.89 × 720 + 100 × 0.10) |
| H100 SXM, 200 GPU-hours (compute only) | $615 at median; $428 at lowest ask; $775 at lowest datacenter ask | $538 | $798 |
| A100 SXM 80GB, 200 GPU-hours (compute only) | $224 at median ($1.122) | $278 | $358 |
Assumptions: 720 hours per month; Vast.ai medians and storage from the 10:00 Europe/Paris snapshot; Runpod card list prices; taxes excluded; Vast.ai egress at the upload rate. Swap in the offer you pick — on Vast.ai, a different host can move the result by tens of percent.
How do interruptible, reserved and serverless options compare?
Vast.ai has more self-serve ways to pay less; Runpod routes discounts through sales. Vast.ai offers on-demand, interruptible (bid) and reserved (prepaid) instances, plus Serverless with no surcharge over the underlying instance cost.
- Interruptible: Vast.ai instances are paused if outbid or if an on-demand renter claims the GPU; docs say “often 50%+ cheaper.” Runpod’s Pods pricing docs checked November 10, 2026 list no interruptible tier.
- Reserved: Vast.ai converts an on-demand instance by prepaying credits; guides say up to 50% off while the API reference says “typically 40%,” and early cancellation claws back the discount plus a 5% fee. Runpod savings plans “are deprecated and can no longer be purchased”; existing plans run to expiry and are non-refundable. Reserved capacity and contract pricing go to Runpod’s enterprise sales team.
- Serverless: Vast.ai bills workers at instance rates with storage and bandwidth billed even while inactive. Runpod bills flex workers per second from start to full stop, rounded up to the nearest second, with a default 5-second idle timeout billed; active-worker discounts go through sales.
Which is more reliable, Vast.ai or Runpod?
Runpod Secure Cloud is the more predictable production tier; both platforms’ cheapest tiers rely on third-party hosts. Runpod describes Secure Cloud as T3/T4 data centers and Community Cloud as peer-to-peer providers with “variable” reliability. Vast.ai labels verified hosts in a Tier 2/3 or ISO 27001 data center as Secure Cloud.
- Datacenter price gap: at 10:00 the lowest Vast.ai datacenter H100 SXM ask was $3.877 per hour (2 offers) versus Runpod Secure at $3.99. Vast.ai returned no datacenter single-GPU offers for H100 NVL, H200, B200, A100 PCIe 80GB or L40S in that pass.
- Community supply: Runpod’s docs state it “is no longer accepting new hosts for Community Cloud.” Vast.ai unverified machines are hidden by default; every machine starts at a 60% reliability score.
- Host responsibility: Runpod’s Terms (last updated March 24, 2026) say Runpod “has no responsibility for the actions of Hosts” and makes no uptime warranties for Community Cloud. Vast.ai docs say it is “only a marketplace” that does not manage the hardware.
- Platform limits: Runpod Pods do not support Docker Compose, UDP connections or Windows (Pods overview). Vast.ai provides Linux Docker instances configured from templates.
How do credits, payments and refunds compare?
Both platforms are prepaid and neither refunds spent credit. The differences are in payment rails and what happens at a zero balance.
| Topic | Vast.ai | Runpod |
|---|---|---|
| Payment | Cards via Stripe; crypto via BitPay and Crypto.com | Cards via Stripe; crypto via integrated processors (KYC first); invoicing over $5,000 with ACH or wire |
| Auto top-up | Autobilling, saved card only | Auto-pay with a saved card, at most once per hour |
| At $0 balance | Instances stop, storage keeps billing; without a card, deletion is scheduled | Pods stop; Pods without a network volume are terminated and their data cannot be recovered |
| Refunds | Spent credits final; unspent balance refundable at Vast.ai’s discretion depending on payment method (Terms, September 1, 2026) | Credits non-refundable and cannot be withdrawn (billing docs) |
Runpod’s zero-balance rule is the harsher one for unattached disks: data on a Pod without a network volume is gone. On Vast.ai, keep a card on file even if you top up with crypto; the billing guide recommends it for exactly that reason.
Which is better for image generation and ComfyUI?
For image generation on 24–32 GB cards, Runpod Community was cheaper on the RTX 4090 and Vast.ai’s cheapest RTX 5090 ask was cheaper than Runpod Community on November 10, 2026. Runpod’s own guide recommends mid-range 16–24 GB cards such as the RTX 4090 for SDXL and Flux.
- RTX 4090 (24 GB): Runpod Community $0.34, Secure $0.89; Vast.ai median $0.536, lowest $0.389.
- RTX 5090 (32 GB): Runpod Community $0.69, Secure $1.19; Vast.ai median $0.743, lowest $0.463.
- Model downloads: repeated checkpoint pulls cost nothing in egress on Runpod; on Vast.ai, check the host’s download rate, since bandwidth is billed both ways.
We did not deploy a ComfyUI template on either platform; check each template’s image and disk size before renting.
Where do the official pages disagree?
- Runpod, storage permanence: the Pods pricing table calls network volumes “Permanent” and the Choose a Pod guide lists them for “long-term storage,” while the same pricing page says “Runpod is not designed for long-term cloud storage” and the billing guide warns network volumes may be terminated if the balance stays at $0.
- Runpod, Community Cloud security: the Pods overview calls Community Cloud “a vetted, secure peer-to-peer system,” while the Terms say Runpod has no responsibility for the actions of Hosts and gives no uptime warranty for Community Cloud.
- Runpod, storage types: the Pods overview says Pods offer three types of storage; the Storage options page lists four, including global volumes at $0.09 per GB per month plus IOPS.
- Runpod, billing cadence: the billing guide says credits are deducted continuously as resources run and, a paragraph later, that “Billing runs every 5 minutes.”
- Runpod, MIG availability: the Choose a Pod guide says RTX PRO 6000 MIG slices are available on Secure Cloud only, while the pricing page’s card data also carries Community Cloud values for them ($0.50 for 24 GB, $1.00 for 48 GB).
- Vast.ai, reserved discount ceiling: guides say “up to 50%,” while the API prepay reference says “Maximum discount is typically 40%.”
Vast.ai or Runpod: which should you buy?
| Buyer | Pick | Why |
|---|---|---|
| A100 80GB training on a budget | Vast.ai | Median ask 19–27% below Runpod Community at 10:00 |
| Checkpointed batch or hyperparameter search | Vast.ai | Interruptible bids; none listed on Runpod Pods |
| Data-heavy inference or dataset pulls | Runpod | No ingress or egress fees |
| RTX 4090 image work | Runpod Community | $0.34 list vs $0.536 Vast.ai median |
| Production needing a data center | Runpod Secure, or a Vast.ai datacenter host | $3.99 vs $3.877 lowest on H100 SXM; Runpod has more listed cards |
| Steady multi-week job, self-serve discount | Vast.ai reserved | Runpod savings plans deprecated |
Key takeaways
- Vast.ai prices GPUs per host offer while Runpod publishes list prices per GPU for Community and Secure Cloud, so a Vast.ai vs Runpod comparison changes by the hour (November 10, 2026).
- At 10:00 Europe/Paris on November 10, 2026, Vast.ai’s median verified A100 SXM 80GB ask was $1.122 per hour, 19% below Runpod’s $1.39 Community Cloud list price.
- At the same time, Runpod Community listed the RTX 4090 at $0.34 per hour versus a $0.536 Vast.ai median verified ask.
- Runpod Pods charge no ingress or egress fees, while Vast.ai bills bandwidth per byte at host-set rates that reached $102.40 per TB on one H100 NVL offer (November 10, 2026).
- Runpod savings plans are deprecated and cannot be purchased, while Vast.ai still sells self-serve reserved prepay discounts with a 5% early-cancellation fee (docs checked November 10, 2026).
- At a $0 balance, Runpod terminates Pods without a network volume and their data cannot be recovered (Runpod billing docs, November 10, 2026).
Our verdict: choose Vast.ai when you will shop offers, tolerate interruptions or want A100 80GB time at marketplace prices; choose Runpod when bandwidth, parked storage or a fixed list price matter more than the lowest possible ask. Price both on the day you rent — neither platform promises today’s numbers tomorrow.
FAQ
Is Vast.ai cheaper than Runpod?
It depends on the GPU. At 10:00 Europe/Paris on November 10, 2026, Vast.ai’s median verified single-GPU ask beat Runpod’s Community Cloud list price on A100 80GB cards ($1.122 vs $1.39 SXM; $0.869 vs $1.19 PCIe). Runpod Community was lower on the RTX 4090 ($0.34 vs $0.536) and H100 SXM ($2.69 vs $3.073), though Vast’s cheapest H100 SXM ask was $2.138.
Which is better for production workloads, Vast.ai or Runpod?
Runpod Secure Cloud is the simpler production pick, because it runs in T3/T4 data centers at one listed price per GPU. Vast.ai has a comparable Secure Cloud label for verified datacenter hosts (Tier 2/3 or ISO 27001), but offers are scarcer: two H100 SXM datacenter offers, lowest $3.877 per hour, versus Runpod Secure at $3.99 (November 10, 2026).
Does Runpod charge for bandwidth like Vast.ai?
No. Runpod Pods docs state there are no fees for data ingress or egress. Vast.ai bills bandwidth per byte at host-set rates: at 10:00 Europe/Paris on November 10, 2026, median upload rates were about $5.33 per TB on verified RTX 4090 offers and $36 per TB on datacenter H100 SXM offers.
Which has cheaper storage, Vast.ai or Runpod?
Runpod’s list storage was cheaper than most Vast.ai host rates checked on November 10, 2026. Runpod charges $0.10 per GB per month for a running volume disk, $0.20 when the Pod is stopped, and $0.07 for network volumes under 1 TB. Vast.ai storage is host-set; medians were $0.20–$0.31 per GB per month on sampled verified A100, L40S, H100 SXM and RTX offers.
Does Runpod have interruptible or spot GPUs?
Not in the current Pods pricing docs. Runpod’s Pods pricing page checked November 10, 2026 lists on-demand pricing and deprecated savings plans that can no longer be bought; reserved capacity goes through sales. Vast.ai offers interruptible bid instances and self-serve reserved prepay discounts on existing on-demand instances.
Can I get a refund from Vast.ai or Runpod?
Mostly no. Runpod billing docs say credits are non-refundable and cannot be withdrawn. Vast.ai Terms (version dated September 1, 2026) say payments for spent credits are final, while unspent balances may be refunded at Vast.ai’s discretion depending on payment method. Vast.ai reserved instances return part of the deposit after a discount clawback plus a 5% fee.
Did Unfair Advantage test Vast.ai and Runpod?
No. We did not open accounts, rent GPUs or run benchmarks on either platform. Vast.ai rates come from its documented search offers API queried on November 10, 2026 (Europe/Paris). Runpod rates come from its public pricing page, marked Updated November 4, 2026 and fetched the same morning, plus both platforms’ docs and Terms.
Sources
Sources are listed as plain URLs, without links, so this page sends no clicks to vendors other than through the disclosed affiliate button. Times are Europe/Paris on November 10, 2026.
- Vast.ai search offers endpoint, https://console.vast.ai/api/v0/bundles/ (verified + rentable + unrented + 1 GPU, on-demand; datacenter pass; snapshot 10:00)
- Vast.ai pricing docs, https://docs.vast.ai/guides/pricing.md and https://docs.vast.ai/guides/instances/pricing.md (checked 09:50–09:51)
- Vast.ai instance types and reserved instances, https://docs.vast.ai/guides/instances/choosing/instance-types.md and https://docs.vast.ai/guides/instances/choosing/reserved-instances.md (checked 09:50)
- Vast.ai finding and renting, https://docs.vast.ai/guides/instances/choosing/find-and-rent.md (tiers, reliability; checked 09:50)
- Vast.ai billing guide and billing FAQ, https://docs.vast.ai/guides/reference/billing.md and https://docs.vast.ai/guides/reference/faq/billing.md (checked 09:51)
- Vast.ai serverless pricing, https://docs.vast.ai/guides/serverless/pricing.md (checked 09:51)
- Vast.ai documentation index, https://docs.vast.ai/llms.txt (prepay “typically 40%”; checked 09:49)
- Vast.ai Terms of Service, https://vast.ai/terms (Version Date: September 1, 2026; refund policy; checked 10:00)
- Runpod pricing page, https://www.runpod.io/pricing (Updated November 4, 2026; Community and Secure Cloud card prices; checked 09:57)
- Runpod Pods pricing, https://docs.runpod.io/pods/pricing.md (storage table, savings plans, account limits; checked 09:58)
- Runpod billing overview, https://docs.runpod.io/accounts-billing/billing.md (payments, zero balance, refunds; checked 09:58)
- Runpod storage options, https://docs.runpod.io/pods/storage/types.md (checked 09:58)
- Runpod Choose a Pod and Pods overview, https://docs.runpod.io/pods/choose-a-pod.md and https://docs.runpod.io/pods/overview.md (checked 09:58)
- Runpod manage Pods, https://docs.runpod.io/pods/manage-pods.md (stop and resume; checked 09:58)
- Runpod Serverless pricing, https://docs.runpod.io/serverless/pricing.md (checked 09:58)
- Runpod Terms of Service, https://www.runpod.io/legal/terms-of-service (Last Updated March 24, 2026; checked 09:57)