GPUBank vs fal
fal is a San Francisco generative-media inference platform, founded in 2021, that sells 1,000+ hosted model APIs, Serverless GPUs for custom apps billed per second of runner lifetime, and dedicated H100 SXM Compute instances. Its H100 serverless list price is $4.50 per GPU-hour, with an as low as rate of $1.89 whose conditions are not published; GPUBank charges $1.85 on demand. fal is built for autoscaling media inference with SOC 2 Type II; GPUBank rents plain instances self-serve at lower list prices.
GPUBank
Tallinn, Estonia
H100 SXM
$1.85 / h
- vCPU, RAM, NVMe, IPv4 and traffic included
- Per-second billing after the health check
- Top up in Bitcoin, USDT and 12+ other coins
vCPU, RAM, NVMe, IPv4 and traffic included
fal
San Francisco, United States · since 2021
H100 SXM · serverless
$4.50 / h
- 1,000+ hosted image, video and audio model APIs billed per output
- Serverless runners that scale from zero, pending time not billed
- B300 and GB200, which GPUBank does not rent
fal
Serverless
United States21 points compared21 of their pages readprices last checked on 29 September 2026page updated 2 November 2026
The verdict
Serverless media inference at fal. Lower list prices and self-serve instances at GPUBank.
fal and GPUBank sell different things. fal runs hosted model APIs and serverless runners that scale from zero, and its dedicated Compute rates are not published. Against fal's serverless list prices, GPUBank costs 34 to 62 % less per GPU-hour on the four GPUs both of us price, although fal's as low as B200 rate of $3.49 is below our $4.15. fal takes cards and ACH, needs team approval for Serverless deploys and publishes no regions; GPUBank takes crypto, launches instances from the console and names four regions.
falGPUBank 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 | fal per hour | GPUBank per hour | Difference | 100 h at fal | 100 h at GPUBank |
|---|---|---|---|---|---|
 B200 SXM180 GB · their serverless price |
$6.25 | $4.15 | −34 % | $625.00 | $415.00 |
 H200 SXM141 GB · their serverless price |
$4.50 | $2.49 | −45 % | $450.00 | $249.00 |
 H100 SXM80 GB · their serverless price |
$4.50 | $1.85 | −59 % | $450.00 | $185.00 |
 RTX PRO 600096 GB · their serverless price |
$2.99 | $1.15 | −62 % | $299.00 | $115.00 |
Where fal publishes no on-demand price for a model, its lowest published serverless price is used, as labelled on the row.
A fine-tune, 40 h on an H100 SXM
fal
$180.00
GPUBank
$74.00
Saves $106.00 at GPUBank (−59 %)
A long-context job, 40 h on an H200
fal
$180.00
GPUBank
$99.60
Saves $80.40 at GPUBank (−45 %)
A full day, 24 h on a B200
fal
$150.00
GPUBank
$99.60
Saves $50.40 at GPUBank (−34 %)
02Their catalogue
Every GPU fal lists.
fal lists 13 GPU models. The 6 below are the ones recorded in our check, each at its lowest public price per GPU-hour. 4 of them are also rented at GPUBank (in blue).
GB200192 GB · serverlessNo price- B300288 GB · serverless$8.50
- B200180 GB · serverless$6.25
- H200141 GB · serverless$4.50
- H100 SXM80 GB · serverless$4.50
- RTX PRO 600096 GB · serverless$2.99
03What the price includes
What appears on the bill besides the GPU.
The same H100 SXM session of 2 h 14 m 03 s, line by line, as each provider describes its billing.
fal
2 h 14 m 03 s on H100 SXM
- GPU · $4.50 / h
- $10.05
- vCPU and RAM
- Included
- Storage
- Not published
- Egress
- Not published
- Public IPv4
- Not published
- Billing granularity
- Per second serverless, per hour Compute
- Minimum charge
- Not published
GPUBank
2 h 14 m 03 s on H100 SXM
- GPU · $1.85 / h
- $4.13
- 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.
fal
Terminated runners free; Compute billed while it exists.
Idle runners kept warm by keep_alive or min_concurrency are billed at the full rate.
GPUBank
Only the persistent disk, $0.03 per GB-month. Compute, IPv4 and scratch cost nothing.
04Where the GPUs run
Where fal runs GPUs is not published.
Countries where fal 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.
Paris
Frankfurt
Ashburn
Singapore
fal0
No country list on its public pages.
GPUBank4
Paris
Frankfurt- Ashburn
Singapore
fal · 0 countriesGPUBank · 4 regions
05Paying and signing up
How each one takes your money.
fal
Payments and sign-up
Payment methods
Cards
Bank transfer
No cryptoMinimum top-up Not published
To start
- 1GitHub or Google OAuth, or SSO/SAML
- 2Prepaid credits
- 3Fal team approval for Serverless deploys
- 4Compute access requested via dashboard
Billing modelPrepaid credits; invoices for higher volume
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.
| fal | GPUBank | |
|---|---|---|
| SOC 2 | Type II | Not published |
| ISO 27001 | Not published | Not published |
| HIPAA | Not published | No |
| GDPR DPA | DPA available | Published DPA, applies automatically |
| Availability target | 99.99% stated on the Serverless page | 99.9 % monthly, per region |
| Status page | status.fal.ai | Public status page |
| Support | Email, Discord; 24/7 priority for 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 fal 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 a 5-row GPU table with List Price and As low as columns, in dollars per hour, without an account.
- 2The Serverless page shows a separate 6-row table including GB200, with no time unit and different figures.
- 3Compute docs list instance specs but send rates to the dashboard or sales.
- 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
21 points, read off both sites.
GPUBank aheadfal aheadEven
| Criterion | GPUBank | fal |
|---|---|---|
| Price and catalogue | ||
| H100 SXM, per GPU-hourConditions behind as low as are not published. A second fal page lists $2.49. | $1.85, all-in | $4.50 serverless list, as low as $1.89 |
| Cheapest GPU-hourOnly six GPUs have a public price. As low as rate is $1.10 with unpublished conditions. | $0.23, RTX 4090 | $2.99 list, RTX PRO 6000 |
| GPU models listedOnly six of the 13 named GPUs have a public price. | 10 | 13Ahead |
| What the GPU price includesCompute instances include 1 TB or 8 TB of SSD. | vCPU, RAM, NVMe, IPv4, traffic | CPU and RAM per machine type |
| Billing | ||
| Billing granularity | Per second | Per second serverless, per hour Compute |
| Minimum charge | None | Not published |
| Stopped instanceIdle runners kept warm by keep_alive or min_concurrency are billed at the full rate. | Disk only | Terminated runners free; Compute billed while it exists |
| Persistent storage | $0.03 per GB-month | Not published |
| Egress traffic | Included | Not published |
| Public IPv4 | Included | Not published |
| Billing modelPurchased credits expire after 365 days and are non-refundable. | Prepaid balance | Prepaid credits; invoices for higher volume |
| Payment and sign-up | ||
| Crypto payments | Bitcoin, USDT, Ethereum and 12+ more | Not mentioned; terms name card and ACH |
| Card payments | No | Cards and ACH in USDAhead |
| Minimum deposit | $25 | Not published |
| Platform | ||
| API and CLI | REST API, CLI, Python SDK | REST APIs, CLI, SDKs in five languages |
| Terraform | Official provider | No provider |
| Serverless GPUs | No | Yes, scale from zeroAhead |
| Managed Kubernetes | No | No |
| Ready-made templatesHosted model APIs, not user templates. | 25 | 1,000+ hosted model endpointsAhead |
| Interruptible or spot | Yes, 40 % off | No |
| GPUs per instance | Up to 8 | Up to 8; multi-node 8x H100 SXMAhead |
| Multi-node interconnect | Not offered | InfiniBand between 8x H100 instancesAhead |
| Network and regions | ||
| Regions | 4 | Not published |
| Countries | France, Germany, US, Singapore | Not published |
| EU regions | Paris, Frankfurt | Not published |
| Trust and support | ||
| Availability targetSLA terms are not published. | 99.9 % monthly | 99.99% stated on the Serverless page |
| Compliance reports | None published | SOC 2 Type IIAhead |
| Public status page | Yes | status.fal.ai |
| Support | Tickets from a $25 balance | Email, Discord; 24/7 priority for enterpriseAhead |
09The right fit
Different needs. A clear choice.
What fal does better
- 1,000+ hosted image, video and audio model APIs billed per output
- Serverless runners that scale from zero, pending time not billed
- B300 and GB200, which GPUBank does not rent
- Multi-node 8x H100 SXM over InfiniBand
- SOC 2 Type II, SSO and invoice billing for enterprise
Choose GPUBank if
- H100 SXM at $1.85 per GPU-hour, against a $4.50 list price
- RTX PRO 6000 at $1.15, against a $2.99 list price
- Self-serve instances, launched from the console
- Published prices for A100, L40S, MI300X, RTX 5090 and RTX 4090
- Four named regions, crypto top-ups, IPv4 and traffic included
Moving from fal takes an afternoon.
fal Serverless apps are Python fal.App classes or standard Docker images, so containers are portable. Generated media on the fal CDN expires after 7 days by default and must be downloaded first.
- 1Download files from /data through the Platform API files endpoints, or copy them from Compute hosts with rsync.
- 2Launch the same GPU on GPUBank with a persistent disk and run your Docker image or HTTP server on it.
- 3Replace fal's queue and autoscaling with your own endpoint, then remove min_concurrency so fal runners stop billing.
# 1. a GPU and a disk on GPUBank gpubank launch --gpu h100-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.
At list price, yes. fal lists the H100 at $4.50 per hour for serverless and advertises as low as $1.89 without publishing the conditions. GPUBank charges $1.85 per GPU-hour for an H100 SXM on demand, with vCPU, RAM, NVMe, IPv4 and traffic included.
No. fal runs serverless runners that scale from zero and bills them per second of runner lifetime, and it hosts 1,000+ model APIs billed per output. GPUBank rents instances only, so for bursty media inference fal can cost less.
fal does not mention crypto. Its terms name payment card and ACH in USD, with invoices for higher-volume customers. GPUBank is topped up in Bitcoin, USDT, Ethereum and more than a dozen other coins.
fal does not publish regions or data center locations. Its Trust Center lists Google Cloud in the United States as an infrastructure subprocessor. GPUBank runs in Paris, Frankfurt, Ashburn and Singapore.
Yes. fal's purchased credits expire 365 days after purchase and are non-refundable, and promotional credits expire in 90 days per the terms. GPUBank publishes no expiry on its prepaid USD balance.
Without. Every price on this page excludes VAT and sales tax, for fal and for GPUBank, so the two columns compare like for like. Tax is added at checkout according to where the buyer is based.
Sources
Every figure about fal was read on its own pages or official documentation. Prices last checked on 29 September 2026.
- 01fal.ai/pricingGPU list and 'as low as' prices ($/h), model API prices
- 02fal.ai/serverlessSecond GPU price table incl. GB200, 99.99% uptime SLA claim, GPU list
- 03fal.ai/Products, SOC 2, usage-based or reserved pricing, A6000 listed
- 04fal.ai/docs/documentation/deployment/machine-typesServerless GPU/CPU machine specs
- 05fal.ai/docs/documentation/serverless/pricingPer-second billing by runner state, GPU count multiplier
- 06fal.ai/docs/documentation/computeH100 SXM instance specs, InfiniBand, per-hour billing
- 07fal.ai/docs/documentation/compute/pricingRates not published
- 08fal.ai/docs/documentation/compute/quickstartEnterprise access, SSH, provisioning
- 09fal.ai/docs/documentation/model-apis/pricingPrepaid credits, output-based billing, errors not billed
- 10fal.ai/docs/documentation/model-apis/faq.mdCredit expiry, invoice billing, volume discounts, concurrency, CDN retention
- 11fal.ai/docs/documentation/serverless/faq.mdIdle billing, GPU types incl. RTX 4090/5090, /data volume
- 12fal.ai/docs/documentation/model-apis/concurrency-limits.md2 to 40 concurrent requests self-serve
- 13fal.ai/docs/documentation/development/use-persistent-storage.md/data volume
- 14fal.ai/docs/documentation/serverlessServerless access approved per account, custom containers
- 15fal.ai/legal/terms-of-serviceLegal entity, card/ACH payment, credit expiry, export control
- 16trust.fal.ai/SOC 2 Type II, subprocessors, HQ
- 17status.fal.ai/Uptime figures
- 18fal.ai/enterpriseSSO, SLA guarantees, private hosting
What we could not verify (9)
- Compute (1xH100-SXM, 8xH100-SXM) hourly rates: not published, dashboard or sales only
- Conditions behind 'As low as' prices (volume, commitment, term): not published
- Time unit of the /serverless price table: not printed; figures conflict with /pricing
- Prices for A100, L40, L40S, RTX 4090, RTX 5090, RTX A6000 and H100 MIG: not on public pages (likely dashboard or Platform API behind a key)
- B300 and GB200 machine specs (CPU/RAM): not in machine-types doc
- Persistent /data storage price, egress and IPv4 policy: not published
- Regions and data center locations: not published (Trust Center lists Google Cloud, United States, as infrastructure subprocessor only)
- Crypto payment: not mentioned; Terms name payment card and ACH in USD only
- Minimum credit purchase amount: not published
In short
GPUBank vs fal, in facts.
The essentials of this comparison in one table, for quick readers and AI assistants. Prices last checked on 29 September 2026.
Read as Markdown- H100 SXM, per GPU-hour
- fal: $4.50 serverless list, as low as $1.89
GPUBank: $1.85, all-in - Cheapest GPU-hour
- fal: $2.99 list, RTX PRO 6000
GPUBank: $0.23, RTX 4090 - Billing granularity
- fal: Per second serverless, per hour Compute
GPUBank: Per second - Stopped instance
- fal: Terminated runners free; Compute billed while it exists
GPUBank: Disk only - Persistent storage
- fal: Not published
GPUBank: $0.03 per GB-month - Egress traffic
- fal: Not published
GPUBank: Included - Crypto payments
- fal: Not mentioned; terms name card and ACH
GPUBank: Bitcoin, USDT, Ethereum and 12+ more - Card payments
- fal: Cards and ACH in USD
GPUBank: No - Minimum deposit
- fal: Not published
GPUBank: $25 - GPU models listed
- fal: 13
GPUBank: 10 - Regions
- fal: Not published
GPUBank: 4 - Serverless GPUs
- fal: Yes, scale from zero
GPUBank: No - Availability target
- fal: 99.99% stated on the Serverless page
GPUBank: 99.9 % monthly - Compliance reports
- fal: SOC 2 Type II
GPUBank: None published - Support
- fal: Email, Discord; 24/7 priority for enterprise
GPUBank: Tickets from a $25 balance - Headquarters
- fal: San Francisco, United States
GPUBank: Tallinn, Estonia
How this page was made
Every price and policy of fal 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. fal, its logo and its website belong to their owner and are shown only to identify the service compared. GPUBank is not affiliated with fal.
