GPUBank vs Koyeb
Koyeb is a French serverless platform from Boulogne-Billancourt that runs containers in microVMs and bills GPU instances by the second, with autoscaling and scale-to-zero; it announced in February 2026 that it is joining Mistral AI. Its H100, form factor not published, costs $2.50 per GPU-hour; GPUBank charges $1.85 for an H100 SXM. Koyeb is built for inference endpoints that sleep when idle; GPUBank rents GPU instances with SSH, takes crypto and charges no plan fee.
GPUBank
Tallinn, Estonia
H200 SXM
$2.49 / h
- vCPU, RAM, NVMe, IPv4 and traffic included
- Per-second billing after the health check
- Top up in Bitcoin, USDT and 12+ other coins
Koyeb
Boulogne-Billancourt, France
H200 SXM · serverless
$3.00 / h
- Serverless GPU Services with autoscaling and scale-to-zero
- Paused Services are not charged
- 99.9 % SLA on Scale and 99.99 % on Enterprise, with service credits
Koyeb
Serverless
25 points compared25 of their pages readprices last checked on 29 September 2026page updated 2 November 2026
The verdict
Scale-to-zero inference and a contractual SLA at Koyeb. Lower GPU prices and no plan fee at GPUBank.
Koyeb suits teams that deploy containerised inference APIs and want autoscaling, scale-to-zero and Terraform or Pulumi automation, with SOC 2 and ISO 27001 listed on its Enterprise plan. For the GPUs both of us rent, GPUBank costs 17 to 56 % less per GPU-hour. New Koyeb users must pay a plan of at least $29 per month, by card only, and most of its GPU types run only in North America. GPUBank has no plan fee, takes crypto, includes traffic and IPv4, and gives SSH access in Europe, the US and Singapore.
01The real cost
The same GPUs, side by side, per hour.
Every GPU model both of us rent, at each provider's own published price for one card, in US dollars, excluding VAT. Prices last checked on 29 September 2026.
| GPU | Koyeb per hour | GPUBank per hour | Difference | 100 h at Koyeb | 100 h at GPUBank |
|---|---|---|---|---|---|
| $5.50 | $4.15 | −25 % | $550.00 | $415.00 | |
| $3.00 | $2.49 | −17 % | $300.00 | $249.00 | |
| $2.20 | $1.15 | −48 % | $220.00 | $115.00 | |
| $2.15 | $0.95 | −56 % | $215.00 | $95.00 | |
| $1.20 | $0.55 | −54 % | $120.00 | $55.00 |
Where Koyeb publishes no on-demand price for a model, its lowest published serverless price is used, as labelled on the row.
An inference box, a month, 720 h, on an L40S
Koyeb
$864.00
GPUBank
$396.00
Saves $468.00 at GPUBank (−54 %)
A long-context job, 40 h on an H200
Koyeb
$120.00
GPUBank
$99.60
Saves $20.40 at GPUBank (−17 %)
A training run, 100 h on an A100 SXM
Koyeb
$215.00
GPUBank
$95.00
Saves $120.00 at GPUBank (−56 %)
02Their catalogue
Every GPU Koyeb lists.
11 GPU models listed by Koyeb, each at its lowest public price per GPU-hour. 5 of them are also rented at GPUBank (in blue).
B200180 GB · serverless$5.50
H200141 GB · serverless$3.00- H10080 GB · serverless$2.50
A100 SXM 80GB80 GB · serverless$2.15
A100 PCIe 80GB80 GB · serverless$1.60
L40S48 GB · serverless$1.20
L424 GB · serverless$0.70
RTX PRO 600096 GB · serverless$2.20
RTX A600048 GB · serverless$0.75- RTX 4000 SFF Ada20 GB · serverless$0.50
- Tenstorrent n300s24 GB · serverlessNo price
03What the price includes
What appears on the bill besides the GPU.
The same H200 SXM session of 2 h 14 m 03 s, line by line, as each provider describes its billing.
Koyeb
2 h 14 m 03 s on H200 SXM
- GPU · $3.00 / h
- $6.70
- vCPU and RAM
- Included
- Storage
- Not published
- Egress
- Metered past an allowance; sources conflict
- Public IPv4
- Shared IP pools; direct IP on Enterprise
- Billing granularity
- Per second
- Minimum charge
- Paid plan from $29 per month
GPUBank
2 h 14 m 03 s on H200 SXM
- GPU · $2.49 / h
- $5.56
- vCPU and RAM
- Included
- Storage
- Local NVMe included; persistent disk $0.03 per GB-month
- Egress
- Included
- Public IPv4
- Included
- Billing granularity
- Per second, from the health check
- Minimum charge
- None
While stopped
What keeps running on the bill.
Koyeb
Paused Services not charged.
Scale-to-zero, in public preview, puts idle GPU Services to sleep.
GPUBank
Only the persistent disk, $0.03 per GB-month. Compute, IPv4 and scratch cost nothing.
04Where the GPUs run
5 countries against 4, and where they are.
Countries where Koyeb says its GPUs run, from its own region list or documentation, against GPUBank's four regions. A pin marks a country, not the exact site.
Koyeb5
United States
Germany
France
Singapore
Japan
GPUBank4
Paris
Frankfurt
Ashburn
Singapore
Koyeb · 5 countriesGPUBank · 4 regions
05Paying and signing up
How each one takes your money.
Koyeb
Payments and sign-up
Payment methods
Minimum top-up None; $29 card pre-authorization hold
To start
- 1Email and password
- 2Card
Billing modelPostpaid, card charged monthly
GPUBank
Payments and sign-up
Payment methods
A prepaid USD balance, topped up in crypto. Every GPU-second is drawn from it.
Minimum top-up $25
To start
- 1Email and password
- 2Top up in crypto
- 3Launch a GPU
Billing modelPrepaid balance
06Trust and support
Certificates, uptime and who answers.
| GPUBank | ||
|---|---|---|
| SOC 2 | Listed, report not public | Not published |
| ISO 27001 | Listed, report not public | Not published |
| HIPAA | Not published | No |
| GDPR DPA | DPA available | Published DPA, applies automatically |
| Availability target | 99.9 % on Scale, 99.99 % on Enterprise | 99.9 % monthly, per region |
| Status page | status.koyeb.com | Public status page |
| Support | Email and chat on Pro; 24x7 on Enterprise | Tickets from a $25 balance |
07Their site, our site
What you see before you pay.
Both sites as a visitor sees them, without an account: the Koyeb pricing page captured on 29 September 2026, and GPUBank's live board. Cookie banners and chat bubbles are closed; nothing else is edited.

- 1The pricing page shows GPU prices per hour, per second and per month without an account, with the H100 at $2.50 per hour.
- 2Every page carries a banner saying Koyeb is joining Mistral AI.
- 3The pricing FAQ gives outbound bandwidth at $0.04/GB while the plan table lists $0.02/GB for EU and US.

- 1Live stock per region and every price, without an account.
- 2Per-hour and per-second prices, with vCPU, RAM, NVMe, IPv4 and traffic in them.
- 3A machine-readable price feed at /prices.json.
08The spec sheet
25 points, read off both sites.
GPUBank aheadKoyeb aheadEven
| Criterion | GPUBank | |
|---|---|---|
| Price and catalogue | ||
| H100 SXM, per GPU-hour | $1.85, all-in | $2.50, H100 (SXM or PCIe not published) |
| Cheapest GPU-hour | $0.23, RTX 4090 | $0.50, RTX 4000 SFF Ada |
| GPU models listed | 10 | 11, including Tenstorrent n300sAhead |
| What the GPU price includes | vCPU, RAM, NVMe, IPv4, traffic | vCPU, RAM, local disk |
| Billing | ||
| Billing granularity | Per second | Per second |
| Minimum chargePro includes $10 of compute per month. The Starter plan is closed to new users. | None | Paid plan from $29 per month |
| Stopped instanceScale-to-zero, in public preview, puts idle GPU Services to sleep. | Disk only | Paused Services not chargedAhead |
| Persistent storageVolumes are a public preview, documented as 1 to 10 GB with no redundancy. | $0.03 per GB-month | Not published |
| Egress trafficPlan table: 1 TB included, then $0.02/GB (EU/US). Docs FAQ: 100 GB free, then $0.04/GB. | Included | Metered past an allowance; sources conflict |
| Public IPv4 | Included | Shared IP pools; direct IP on Enterprise |
| Billing model | Prepaid balance | Postpaid, card charged monthly |
| Payment and sign-up | ||
| Crypto payments | Bitcoin, USDT, Ethereum and 12+ more | Not accepted |
| Card payments | No | Visa, Mastercard, Amex, Diners, Discover, UnionPayAhead |
| Minimum deposit | $25 | None; $29 card pre-authorization hold |
| Platform | ||
| API and CLI | REST API, CLI, Python SDK | REST API, CLI, Python and JavaScript SDKs |
| Terraform | Official provider | Official provider |
| Serverless GPUs | No | Yes, with scale-to-zeroAhead |
| Managed Kubernetes | No | Not published |
| Ready-made templatesA one-click apps catalogue exists; its size is not published. | 25 | Not published |
| Interruptible or spot | Yes, 40 % off | No |
| GPUs per instanceThe catalog API marks all 8-GPU types as RESTRICTED. | Up to 8 | Up to 8 (A100, A100 SXM, H100, H200) |
| Multi-node interconnect | Not offered | Not published |
| Network and regions | ||
| Regions | 4 | 11 listed; most GPU types in North AmericaAhead |
| Countries | France, Germany, US, Singapore | US, Germany, France, Singapore, JapanAhead |
| EU regionsThe catalog API lists only the RTX 4000 SFF Ada in Frankfurt. | Paris, Frankfurt | Frankfurt and Paris, few GPU types |
| Trust and support | ||
| Availability targetScale costs $299 per month and Enterprise starts at $1000 per month. | 99.9 % monthly | 99.9 % on Scale, 99.99 % on Enterprise |
| Compliance reportsReport types sit behind a Trust Center access request. | None published | SOC 2 and ISO 27001 listed on EnterpriseAhead |
| Public status page | Yes | status.koyeb.com |
| Support | Tickets from a $25 balance | Email and chat on Pro; 24x7 on EnterpriseAhead |
09The right fit
Different needs. A clear choice.
What Koyeb does better
- Serverless GPU Services with autoscaling and scale-to-zero
- Paused Services are not charged
- 99.9 % SLA on Scale and 99.99 % on Enterprise, with service credits
- SOC 2 and ISO 27001 listed on Enterprise, public GDPR DPA
- Pulumi provider, Sandboxes SDK and Tenstorrent accelerators
Choose GPUBank if
- H100 SXM at $1.85, against $2.50 for an H100 of unstated form factor
- H200 at $2.49 and B200 at $4.15, against $3.00 and $5.50
- No monthly plan fee, where Koyeb Pro starts at $29 per month
- Crypto top-ups with no card hold or validation delay
- All traffic and a public IPv4 included, and SSH on every instance
Moving from Koyeb takes an afternoon.
Koyeb workloads are container images, so the same image runs on GPUBank. Volume data must be copied out from inside the running Service, since no snapshot export is documented.
- 1Launch the same GPU on GPUBank with a persistent disk in the region you need.
- 2Copy volume data out of the running Koyeb Service to object storage, then pull it onto the GPUBank disk.
- 3Run the same container image on GPUBank, then pause or delete the Koyeb Services.
# 1. a GPU and a disk on GPUBank gpubank launch --gpu h200-sxm --region eu-paris --disk data # 2. copy your data from the old machine rsync -azP old-host:/workspace/ root@<new-ip>:/workspace/ # 3. run the same image or template ssh root@<new-ip> 'cd /workspace && ./start.sh'
A few things you might be wondering.
Yes, on every GPU both of us list. Koyeb charges $2.50 per GPU-hour for an H100 whose form factor it does not publish, $3.00 for an H200 and $1.20 for an L40S. GPUBank charges $1.85 for an H100 SXM, $2.49 for an H200 and $0.55 for an L40S. An L40S for a 720-hour month costs $864.00 on Koyeb and $396.00 on GPUBank.
Yes, for new users. Koyeb now requires a paid plan: Pro at $29 per month, which includes $10 of compute, or Scale at $299 per month, which includes $100. GPUBank has no plan fee and bills only what you run, from a prepaid balance.
No. Koyeb's pricing FAQ lists card brands only, charged monthly through Stripe after a $29 pre-authorization hold. GPUBank is paid from a USD balance topped up in Bitcoin, USDT, Ethereum and more than a dozen other coins, with no card.
Koyeb. Its GPU Services can scale to zero after an idle period, and paused Services are not charged. GPUBank has no serverless product; you stop an instance yourself and then pay only for its disk, at $0.03 per GB-month.
Koyeb announced a definitive agreement to join Mistral AI on 17 February 2026, subject to closing conditions. The platform keeps operating and billing directly and will gradually become part of Mistral Compute.
Without. Every price on this page excludes VAT and sales tax, for Koyeb and for GPUBank, so the two columns compare like for like. Tax is added at checkout according to where the buyer is based.
Sources
Every figure about Koyeb was read on its own pages or official documentation. Prices last checked on 29 September 2026.
- 01koyeb.com/pricingGPU prices and specs, plans, egress, SLA tiers, compliance claims
- 02koyeb.com/docs/reference/instancesPer-second/hour/month prices, instance specs (docs variant)
- 03app.koyeb.com/v1/catalog/instances?limit=100GPU prices per second/hour, regions, availability status
- 04app.koyeb.com/v1/catalog/regions?limit=100Region and datacenter list
- 05koyeb.com/docs/reference/regionsCore regions
- 06koyeb.com/docs/faqs/pricingBilling, card verification, payment methods, currency, taxes
- 07koyeb.com/docs/reference/volumesPersistent storage limits
- 08koyeb.com/docs/reference/snapshotsSnapshots
- 09koyeb.com/docs/run-and-scale/scale-to-zeroIdle behaviour for GPU services
- 10koyeb.com/docs/reference/accountsSign-up methods
- 11koyeb.com/docs/legal/slaSLA mechanics, service credits
- 12koyeb.com/docs/legal/data-processing-agreementGDPR DPA, subcontractors
- 13koyeb.com/docs/legal/termsLegal entity, address, governing law
- 14koyeb.com/docs/integrations/infrastructure-as-code/terraformOfficial Terraform provider
- 15koyeb.com/docs/integrations/infrastructure-as-code/pulumiPulumi provider
- 16koyeb.com/docs/reference/apiREST API, OpenAPI
- 17koyeb.com/docs/sandboxes/sandbox-quickstartPython and JavaScript SDKs
- 18koyeb.com/docs/hardware/nvidia-a100A100 PCIe vs SXM distinction
What we could not verify (10)
- H100 form factor (SXM vs PCIe) not published on pricing, docs or catalog API; H100 rows recorded as model 'other', so H100 SXM price is null.
- Founding year not stated as such; the about page only says the team 'set out' in 2021.
- Volume and snapshot prices not published on pages read.
- Saving-plan (reserved) prices and terms not published (up to 50% off, contact sales).
- Tenstorrent n300s price not published (request access).
- SOC 2 report type and ISO 27001 scope not verifiable: Trust Center requires an access request.
- Egress: pricing page plan table and docs pricing FAQ give conflicting allowances and rates.
- GPU interconnect (NVLink/NVSwitch) for multi-GPU instances not published.
- Direct SSH on NVIDIA GPU instances not documented; SSH via TCP proxy is mentioned only for Tenstorrent.
- Crypto, PayPal and wire transfer not mentioned; the FAQ lists card brands only.
In short
GPUBank vs Koyeb, in facts.
The essentials of this comparison in one table, for quick readers and AI assistants. Prices last checked on 29 September 2026.
- H100 SXM, per GPU-hour
- Koyeb: $2.50, H100 (SXM or PCIe not published)
GPUBank: $1.85, all-in - Cheapest GPU-hour
- Koyeb: $0.50, RTX 4000 SFF Ada
GPUBank: $0.23, RTX 4090 - Billing granularity
- Koyeb: Per second
GPUBank: Per second - Stopped instance
- Koyeb: Paused Services not charged
GPUBank: Disk only - Persistent storage
- Koyeb: Not published
GPUBank: $0.03 per GB-month - Egress traffic
- Koyeb: Metered past an allowance; sources conflict
GPUBank: Included - Crypto payments
- Koyeb: Not accepted
GPUBank: Bitcoin, USDT, Ethereum and 12+ more - Card payments
- Koyeb: Visa, Mastercard, Amex, Diners, Discover, UnionPay
GPUBank: No - Minimum deposit
- Koyeb: None; $29 card pre-authorization hold
GPUBank: $25 - GPU models listed
- Koyeb: 11, including Tenstorrent n300s
GPUBank: 10 - Regions
- Koyeb: 11 listed; most GPU types in North America
GPUBank: 4 - Serverless GPUs
- Koyeb: Yes, with scale-to-zero
GPUBank: No - Availability target
- Koyeb: 99.9 % on Scale, 99.99 % on Enterprise
GPUBank: 99.9 % monthly - Compliance reports
- Koyeb: SOC 2 and ISO 27001 listed on Enterprise
GPUBank: None published - Support
- Koyeb: Email and chat on Pro; 24x7 on Enterprise
GPUBank: Tickets from a $25 balance - Headquarters
- Koyeb: Boulogne-Billancourt, France
GPUBank: Tallinn, Estonia
How this page was made
Every price and policy of Koyeb was read on its own website, documentation or public API on 29 September 2026, with the address of each page listed above. A value we could not read on a primary source is marked "Not published" rather than estimated. GPUBank figures come from this site and its terms. Prices change: check both providers before you buy. Koyeb, its logo and its website belong to their owner and are shown only to identify the service compared. GPUBank is not affiliated with Koyeb.