Operator · San Francisco, CA · founded 2021

fal

San Francisco generative-media inference platform: 1,000+ hosted model APIs billed per output, Serverless for custom apps billed per second of runner lifetime, and dedicated H100 SXM Compute instances.

GPU types6with a current offer

Current offers3031 readings on file · last 45 days

H100 from—per GPU-hour, on-demand

Entry price—cheapest GPU-hour listed

Countries1regions not published

Trust Score v0.1

87/100

Company & track record100
Infrastructure control100
Pricing & billing transparency65
Reliability signals100
Security & compliance73
Customer signal—

Scored Sep 29, 2026 from the provider's own pages. Dimensions not yet measured are excluded from the denominator. Criteria ↓

30 current offers →fal.ai ↗

Trust Score · 5 of 6 dimensions measured

The score, criterion by criterion

Each point comes from something the provider publishes, with the page it was read on and the date. Anything not disclosed earns zero.

Company & track record

weight 15%

100 / 100

Founding year published	2021fal.ai · Sep 29, 2026	4 / 4
Headquarters published	Features & Labels, Inc.fal.ai · Sep 29, 2026	3 / 3
Legal entity named	Features & Labels, Inc.fal.ai · Sep 29, 2026	4 / 4
Funding or ownership disclosed	Private; 'backed by the best Silicon Valley investors and angels' (no amounts on the pages read)fal.ai · Sep 29, 2026	4 / 4

Infrastructure control

weight 15%

100 / 100

Capacity model published	operates or rents dedicated capacityfal.ai · Sep 29, 2026	10 / 10
Regions published	1 regions or countriesfal.ai · Sep 29, 2026	5 / 5

Pricing & billing transparency

weight 20%

65 / 100

Public prices without an account	yes (read on the pricing page)fal.ai · Sep 29, 2026	5 / 5
Billing granularity published	mixedfal.ai · Sep 29, 2026	4 / 4
Egress policy published	not disclosedfal.ai · Sep 29, 2026	0 / 4
Stopped-instance billing published	Serverless: nothing billed once a runner is TERMINATED; IDLE runners (keep_alive, min_concurrency), setup() and draining are billed at the full per-second rate. Compute: billed per hour at a fixed rate while the instance exists, regardless of utilization.fal.ai · Sep 29, 2026	4 / 4
Public IPv4 policy published	not disclosedfal.ai · Sep 29, 2026	0 / 3

Reliability signals

weight 20%

100 / 100

Status page	https://status.fal.ai/fal.ai · Sep 29, 2026	6 / 6
SLA published	'99.99% uptime SLA' stated on the Serverless product page; enterprise page mentions SLA guarantees (terms not published)fal.ai · Sep 29, 2026	6 / 6
SLA with a numeric uptime commitment	'99.99% uptime SLA' stated on the Serverless product page; enterprise page mentions SLA guarantees (terms not published)fal.ai · Sep 29, 2026	3 / 3
Uptime history published	Status page, 90 days: API 99.98%, Model API 99.97%, Serverless API 100.0% (read 2026-09-29, with an incident open that day)fal.ai · Sep 29, 2026	5 / 5
Observed availability (30 days)	not measured yet	—

Security & compliance

weight 15%

73 / 100

SOC 2	Type IIfal.ai · Sep 29, 2026	8 / 8
ISO 27001	not disclosedfal.ai · Sep 29, 2026	0 / 4
HIPAA or GDPR DPA	DPA referenced in subprocessor notices; text not publicfal.ai · Sep 29, 2026	3 / 3

Customer signal

weight 15%

not measured

Aggregated public reviews

not collected yet

—

provisional v0.1: evidence from the provider's own pages; dimensions not yet measured are excluded from the denominatorFull method →

Prices by GPU · 6 GPU types with a current offer

What fal charges, against the market

Cheapest current on-demand true cost here, next to the market median for the same GPU. Negative means cheaper than the market.

GPU	Cheapest here	Market median	vs market	Current offers	Modes
H100	no on-demand	$3.17	—	2	serverlessreserved	H100 prices
H200	no on-demand	$4.80	—	2	serverlessreserved	H200 prices
RTX PRO 6000	no on-demand	$2.15	—	2	serverlessreserved	RTX PRO 6000 prices
B200	no on-demand	$7.50	—	2	serverlessreserved	B200 prices
B300	no on-demand	$8.99	—	2	serverlessreserved	B300 prices
GB200	no on-demand	$8.00	—	2	serverlessreserved	GB200 prices

All offers · 31 readings on file

Every fal offer, with its receipt

Current offers (last 45 days) ranked by true cost, all rental modes, each with its read date. Older readings stay listed below, in grey.

GPU	Configuration	Mode	Region	Stock	List	True cost	Read
RTX PRO 6000	1× RTX PRO 6000	Reserved	—	Stock unknown	$1.99	$1.99	00:32 UTCfresh	Receipt
H100SXM	1× H100 SXM	Reserved	—	Stock unknown	$2.49	$2.49	00:32 UTCfresh	Receipt
H200	1× H200	Reserved	—	Stock unknown	$2.99	$2.99	00:32 UTCfresh	Receipt
RTX PRO 6000	1× RTX PRO 6000	Serverless	—	Stock unknown	$4.00	$4.00	00:32 UTCfresh	Receipt
H100SXM	1× H100 SXM	Serverless	—	Stock unknown	$4.50	$4.50	00:32 UTCfresh	Receipt
B200	1× B200	Reserved	—	Stock unknown	$5.49	$5.49	00:32 UTCfresh	Receipt
GB200	1× GB200	Reserved	—	Stock unknown	$5.89	$5.89	00:32 UTCfresh	Receipt
B300	1× B300	Reserved	—	Stock unknown	$5.99	$5.99	00:32 UTCfresh	Receipt
H200	1× H200	Serverless	—	Stock unknown	$6.00	$6.00	00:32 UTCfresh	Receipt
B200	1× B200	Serverless	—	Stock unknown	$7.99	$7.99	00:32 UTCfresh	Receipt
GB200	1× GB200	Serverless	—	Stock unknown	$9.99	$9.99	00:32 UTCfresh	Receipt
B300	1× B300	Serverless	—	Stock unknown	$12.99	$12.99	00:32 UTCfresh	Receipt

12 fresh offers · prices per GPU-hour · true cost includes the add-ons the provider publishesOpen in the comparison tool →

True cost & billing · read Sep 29, 2026

What the hourly price does not say

Billing granularity, minimums, what a stopped machine costs, traffic, addresses, deposits and identity checks, as published by the provider.

Billing

Billing granularity: mixed
Minimum charge: not published
Stopped instance: Serverless: nothing billed once a runner is TERMINATED; IDLE runners (keep_alive, min_concurrency), setup() and draining are billed at the full per-second rate. Compute: billed per hour at a fixed rate while the instance exists, regardless of utilization.
Egress: not published
Public IPv4: not published
Billing model: both
Minimum deposit: not published
Free credits: Free credits and coupons granted case by case; the FAQ gives expiry of 1 week to 1 year, the Terms say promotional credits expire in 90 days. Purchased credits expire 365 days after purchase and are non-refundable.

Account

Payment: card · invoice · crypto (not published)
Sign-up: GitHub or Google OAuth, or SSO/SAML, prepaid credits (card or ACH) before use, fal team approval for Serverless deploys, Compute access requested via dashboard (marked Enterprise)
Identity check: not published
Support: email ([email protected]), discord, dashboard
Enterprise support: not published
SLA: '99.99% uptime SLA' stated on the Serverless product page; enterprise page mentions SLA guarantees (terms not published)
Status page: status.fal.ai ↗
Published uptime: Status page, 90 days: API 99.98%, Model API 99.97%, Serverless API 100.0% (read 2026-09-29, with an incident open that day)

Infrastructure

Regions: —
Countries: —
Platform: API, CLI, serverless, SSH, custom Docker images, persistent volumesSDK: Python (fal, fal-client), JavaScript/TypeScript (@fal-ai/client), Swift, Kotlin/Java, DartKubernetes: noInterconnect: InfiniBand between 8xH100-SXM Compute instances in the same sector; NVLink 4.0 900 GB/s on H100 (machine-types doc)Clusters: Multi-node 8xH100-SXM in one sector over InfiniBand; dedicated clusters for research labs via sales
Storage: network · not published · Persistent /data volume shared across all runners of the account (local NVMe, datacenter cache, global object store).ephemeral · included · Compute instances include 1 TB SSD (1xH100-SXM) or 8 TB SSD (8xH100-SXM).object · not published · Generated media on fal CDN, kept at least 7 days by default.
Migration out: Serverless apps are Python fal.App classes or standard Docker images/HTTP servers (Direct Server Mode), so containers are portable; files on /data can be downloaded through the Platform API files endpoints. Compute instances are Ubuntu SSH hosts, data leaves via scp/rsync. Generated media on fal CDN expires (7 days default) and must be downloaded.

Security & compliance

SOC 2: Type II
ISO 27001: not published
HIPAA: not published
GDPR / DPA: DPA referenced in subprocessor notices; text not public
Other: Penetration report and CAIQ v4.1.0 in Trust Center (access on request)
Legal entity: Features & Labels, Inc.
Funding: Private; 'backed by the best Silicon Valley investors and angels' (no amounts on the pages read)

Read on the provider's own pages · 22 pages read · facts marked "not published" were looked for and not foundPages read ↓

Assessment · documents read Sep 29, 2026

fal, in short

What stood out when we read the documents: strengths, weaknesses, who it suits, and what we could not find.

Strengths

1,000+ hosted generative media model endpoints billed per output; server errors (HTTP 500+) and queue time not charged
Serverless for custom models: per-second billing, pending time and image pulls not billed, autoscaling from zero, custom Docker images from private registries
Public list prices for B300, GB200, B200, H200, H100 and RTX PRO 6000
SOC 2 Type II report (latest announced 27 April 2026) in Trust Center
Migration guides from Replicate, Modal, RunPod and Baseten
Public status page with per-component 90-day uptime

Weaknesses

Two public pages give different GPU prices on the same date (H100 'as low as' 1.89 USD/h on /pricing vs 2.49 on /serverless; B300 list 8.50/h vs 12.99)
Serverless deploys require manual approval by the fal team; Compute access is requested via the dashboard and marked Enterprise
Compute hourly rates not published ('Contact the fal team or check the dashboard')
Idle runners (keep_alive, min_concurrency) and setup() time billed at the full rate
Model API concurrency starts at 2 requests, scales with purchase history to 40 self-serve
Purchased credits expire after 365 days and are non-refundable
No published regions, storage price, egress or IPv4 policy
Compute limited to 1x and 8x H100 SXM

Ideal for

Teams building image, video and audio generation products who want per-output APIs for hosted models
Teams deploying custom diffusion or media models that need autoscaling serverless GPUs and accept per-second runner billing
Enterprises needing SOC 2 Type II, SSO and invoice billing

Not published

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
Identity/KYC requirements: not documented
Terms contain US export-control representations
Minimum credit purchase amount: not published

Observations

fal.ai/pricing shows a 5-row GPU table (List Price and 'As low as', $/h) without an account, plus per-output model prices

fal.ai/serverless shows a separate 6-row table including GB200, with no time unit and different figures

Compute docs list instance specs but defer rates to the dashboard or sales

Headline 'as low as $1.89/hr for H100' on the pricing page

Pages read

fal.ai/pricing — GPU list and 'as low as' prices ($/h), model API prices
fal.ai/serverless — Second GPU price table incl. GB200, 99.99% uptime SLA claim, GPU list
fal.ai/ — Products, SOC 2, usage-based or reserved pricing, A6000 listed
fal.ai/docs/documentation/deployment/machine-types — Serverless GPU/CPU machine specs
fal.ai/docs/documentation/serverless/pricing — Per-second billing by runner state, GPU count multiplier
fal.ai/docs/documentation/compute — H100 SXM instance specs, InfiniBand, per-hour billing
fal.ai/docs/documentation/compute/pricing — Rates not published
fal.ai/docs/documentation/compute/quickstart — Enterprise access, SSH, provisioning
fal.ai/docs/documentation/model-apis/pricing — Prepaid credits, output-based billing, errors not billed
fal.ai/docs/documentation/model-apis/faq.md — Credit expiry, invoice billing, volume discounts, concurrency, CDN retention
fal.ai/docs/documentation/serverless/faq.md — Idle billing, GPU types incl. RTX 4090/5090, /data volume
fal.ai/docs/documentation/model-apis/concurrency-limits.md — 2 to 40 concurrent requests self-serve
fal.ai/docs/documentation/development/use-persistent-storage.md — /data volume
fal.ai/docs/documentation/serverless — Serverless access approved per account, custom containers
fal.ai/docs/documentation/setting-up/accounts-and-identity — Sign-in methods, API key scopes
fal.ai/legal/terms-of-service — Legal entity, card/ACH payment, credit expiry, export control
trust.fal.ai/ — SOC 2 Type II, subprocessors, HQ
status.fal.ai/ — Uptime figures
fal.ai/enterprise — SSO, SLA guarantees, private hosting
fal.ai/about — Founded 2021, founders, HQ
fal.ai/docs/documentation/model-apis/support — Support channels
fal.ai/docs/documentation — Compute vs Serverless billing summary

Questions about fal

How does fal bill?

Billing granularity: mixed. Stopped instance: Serverless: nothing billed once a runner is TERMINATED; IDLE runners (keep_alive, min_concurrency), setup() and draining are billed at the full per-second rate. Compute: billed per hour at a fixed rate while the instance exists, regardless of utilization.

How trustworthy is fal?

Its Trust Score is 87/100 (v0.1), read from its own documents. Strong on company & track record, infrastructure control, reliability signals. Weaker on pricing & billing transparency. The score measures transparency and published guarantees, not customer satisfaction; every criterion and its source is listed above.

Where are fal's data centres?

1 countries: —.

How do you pay fal, and are there credits?

Payment by card, invoice, crypto. Free credits: Free credits and coupons granted case by case; the FAQ gives expiry of 1 week to 1 year, the Terms say promotional credits expire in 90 days. Purchased credits expire 365 days after purchase and are non-refundable..