AI & Agents AI & Developer Infrastructure

Fal's $16B Round: Best Image & Video AI APIs 2026

Fal raised $1.3 billion at a $16 billion valuation on November 3. We compare it with Replicate, Together AI, Runware, WaveSpeedAI and Modal on published prices.

By Arthur R. Published November 4, 2026 Updated November 4, 2026 11 min read

Illustration: Innvesti (AI-generated)

Quick answer

Fal is the broadest hosted API for image and video models and the safest default after its $1.3 billion Series F, but it is not the cheapest for every model. Runware, WaveSpeedAI and Together AI undercut it on several popular image models, and Modal is better if you run your own code.

Key takeaways

Fal announced a $1.3 billion Series F at a $16 billion valuation on November 3, 2026, double the $8 billion it was valued at in April (TechCrunch); Bessemer led, with Nvidia, Sound Ventures and T. Rowe Price in the round.
Prices for the same model differ by platform. Nano Banana Pro costs $0.15 per image on fal, $0.14 on WaveSpeedAI and $0.134 on Together AI; Nano Banana 2 costs $0.08 on fal against $0.05 on Together AI and $0.045 on WaveSpeedAI (vendor pricing pages, checked November 4, 2026).
Fal bills model APIs per output and serverless GPUs per second. It says it does not charge for server errors or queue wait time.
On an H100, list prices run from about $3.95 per hour (Runware, Modal) through $4.50 (fal) to $5.49 (Replicate, Together AI), though fal lists discounted rates as low as $2.49.
Replicate is now part of Cloudflare, which matters if you want your model API near Workers. It still bills by run time or per output.

$1.3B

Fal Series F, announced November 3, 2026 (TechCrunch)

$16B

Fal valuation, double its April 2026 valuation (TechCrunch)

$1.65B

Combined Series E and Series F raised by fal (TechCrunch)

Fal, the hosted image, video and audio model platform, raised $1.3 billion in a Series F at a $16 billion valuation on November 3, 2026, but it is not the cheapest place to run every model. On the vendor pricing pages we checked, Nano Banana 2 costs $0.08 per image on fal and $0.045 to $0.05 on two competitors. The round, led by Bessemer with Nvidia, Sound Ventures and T. Rowe Price, doubled the company’s $8 billion valuation from April, according to TechCrunch, which puts fal’s combined Series E and F at $1.65 billion.

For a developer choosing an image or video API, the price list matters more than the funding. This report compares six platforms on what their own pricing pages said on November 4, 2026.

What the round does and does not tell you

A $16 billion valuation says investors expect demand for generative-media inference to keep growing. It says nothing about the price of the model you plan to call. TechCrunch notes that fal’s competitors Baseten, Fireworks AI and Modal also run on thin gross margins because of compute costs. Prices on shared models can therefore move quickly, so recheck the page before committing a budget.

The same model, different prices

Reading the pricing pages side by side, the clearest finding is how much the same model varies. Nano Banana Pro is $0.15 per image on fal, $0.14 on WaveSpeedAI and $0.134 on Together AI. Nano Banana 2 is $0.08 on fal, $0.05 on Together AI and $0.045 on WaveSpeedAI. For a high-volume workload, a gap of that size matters more than any funding headline. One caveat: listed prices are starting points at the lowest resolution or default settings, so confirm the configuration you need.

Hosted catalog or your own code

This list covers two different kinds of product. Fal, Replicate, Runware, WaveSpeedAI and Together AI sell you a model behind an API. Modal sells compute for code you write. On H100 list prices, Modal and Runware are about $3.95 per hour, fal $4.50, and Replicate and Together AI $5.49. Fal lists lower rates, down to $2.49 per hour, without stating the conditions on the page. If you only need to call a model, price per output is the right unit; if you run custom weights, price per GPU-second is.

Billing details that change real cost

Read what is not billed. Fal says it does not charge for server errors or queue wait time. Runware says it bills only successful requests and does not bill rate-limited ones. Replicate bills private models for all online time, including setup and idle time. Those clauses can outweigh a difference of a few tenths of a cent per image.

1. fal

Best all-round hosted API for image, video and audio models

Fal is the default pick for teams that want one hosted API for current image and video models without managing GPUs. Its model APIs are billed per output, such as Kling Video v3 Image to Video [Pro] at $0.14 per second, Nano Banana Pro at $0.15 per image, FLUX 2 at $0.012 per megapixel and FLUX.1 [schnell] at $0.003 per megapixel, plus serverless GPUs billed per second while active. Its pricing page says it does not charge for server errors or queue wait time, and lists H100 at $4.50 per hour with discounted rates down to $2.49. The new money is mostly a signal of staying power: TechCrunch reports the $1.3 billion Series F, led by Bessemer with Nvidia, Sound Ventures and T. Rowe Price, doubled its valuation from $8 billion in April. The trade-off: the pricing page mentions no free credits, and on several third-party image models fal is not the lowest price.

Pros

Wide catalog of image, video and audio models behind one API, with per-output pricing
No charge for server errors or queue wait time (per its pricing page)
Serverless GPUs for custom models, billed per second

Cons

Not the cheapest on every shared model (Nano Banana 2 is $0.08 versus $0.045 to $0.05 elsewhere)
No free credits mentioned on the pricing page
Mixed units (per second, per image, per megapixel, per token) make cost forecasting harder

2. Replicate

Best for a huge community model catalog, now part of Cloudflare

Replicate is the easiest way to run a long tail of open and community models through a simple API. Public models are mostly billed by run time at hardware rates, such as an H100 at $0.001525 per second ($5.49 per hour) and an L40S at $3.51 per hour. Some are billed per output, for example flux-schnell at $3.00 per thousand images and wan-2.1-i2v-720p at $0.25 per second of video. Cloudflare announced it would acquire Replicate in November 2025, and Cloudflare's FY2025 10-K records the acquisition among Q4 2025 activity; the exact closing date is not confirmed in the sources we checked. The catch is run-time billing: private models are billed for all online time, including setup and idle time, which can surprise teams used to per-output prices.

Pros

Very large catalog of community and open models behind a simple API
Per-output pricing on many popular models, per-second hardware rates on the rest
Backed by Cloudflare following the acquisition

Cons

Private deployments bill for idle and setup time
Run-time billing makes per-image cost vary with hardware and settings
Higher H100 rate ($5.49 per hour) than fal, Runware or Modal at list price

3. Runware

Best for low per-image prices and cheap GPUs

Runware is built around low-cost inference and publishes its rates. Its rate sheet lists FLUX 3 Image Klein at $0.003 per image (512x512, 16 steps) and P-Video-2 at $0.01 per second at 480p, and says the sheet covers 300-plus models with published rates. Serverless GPUs are billed per second, with H100 at $3.95 per hour and reserved rates as low as $2.45. It says you are charged only for successful requests and rate-limited requests are not billed. New users with a business email get $2 in free credits.

Pros

Published per-output prices across 300-plus models
Charges only for successful requests; rate-limited requests are not billed
Lower H100 list price than fal ($3.95 versus $4.50 per hour)

Cons

Smaller free credit ($2, business email only)
Prices depend on parameters, so the page recommends estimating in its Playground
Catalog and quality not tested here; the rate sheet is a price list, not a benchmark

4. WaveSpeedAI

Best for cheap image and video generation through one pay-per-use account

WaveSpeedAI sells pay-per-use access to image, video and language models with no monthly fee; prepaid credits are spent as you generate. Its starting prices include Z-Image Turbo at $0.005 per image, Nano Banana 2 at $0.045, Nano Banana Pro at $0.14 and Kling 3.0 Std video at $0.084 per second. Eligible new accounts get $1 of free credit without a card, and account levels from Bronze to Ultra are set by top-up size. Listed prices are for the lowest resolution or quality, so real costs run higher. It does not publish a GPU-hour rate for running your own code.

Pros

Low starting prices on popular image models
No monthly fee; $1 free credit for eligible new accounts
Image, video and language models under one account

Cons

Listed prices are minimums at the lowest resolution or quality
Some premium models may not work with trial credits
No self-serve GPU product for custom code on the pricing page

5. Together AI

Best if you also run language models and dedicated GPUs

Together AI fits best when image generation is one workload among several, including language models and dedicated endpoints. Its pricing page lists FLUX.2 [dev] at $0.0154 per image, GPT Image 2 at $0.053, Nano Banana 2 at $0.05 and Nano Banana Pro at $0.134, with video models priced per video, such as Seedance 2.0 at $0.16. Dedicated Inference on an HGX H100 is $5.49 per hour and an HGX B200 is $8.99. It undercuts fal on Nano Banana 2 and Nano Banana Pro, but per-video pricing at the lowest settings is hard to compare with per-second rates elsewhere.

Pros

Lower price than fal on Nano Banana 2 and Nano Banana Pro
Language models and dedicated GPUs on the same platform
Per-video pricing is simple for short clips

Cons

Higher H100 rate ($5.49 per hour) than fal, Runware or Modal at list price
Video prices shown at the lowest resolution or duration settings
Free credit amount not stated on the pricing page

6. Modal

Best for running your own model code serverlessly

Modal is not a model catalog. It is a serverless compute platform where you ship your own code and weights, which suits teams that outgrow hosted model APIs and want full control. Its published rates are per second: H100 at $0.001097 (about $3.95 per hour), B200 at $0.001736, A100 80 GB at $0.000694 and L40S at $0.000542. The Starter plan is $0 plus compute with $30 per month in free credits, and the Team plan is $250 per month plus compute with $100 in credits. TechCrunch names Modal among fal's competitors. It offers no one-line call to a hosted video model, so it suits engineering teams, not prompt-and-go use.

Pros

Competitive per-second GPU rates (H100 about $3.95 per hour) and $30 monthly free credits on Starter
Full control over code, weights and scaling
Clear plan tiers up to enterprise

Cons

No hosted catalog of third-party image and video models
You write and maintain the inference code
Cost depends on your own utilization and cold starts

Fal's $16B Round: Best Image & Video AI APIs 2026 — comparison
Platform	Billing model	Nano Banana Pro (per image)	H100 list price (per hour)	Free credit (pricing page)
fal	Per output; GPUs per second	$0.15	$4.50 (as low as $2.49)	Not mentioned
Replicate	Run time or per output	Not listed on pricing page	$5.49	Not mentioned
Runware	Per output; GPUs per second	Not listed in rate excerpt	$3.95	$2 (business email)
WaveSpeedAI	Prepaid credits, per use	$0.14	Not listed	$1 (eligible new accounts)
Together AI	Per image or video; GPUs per hour	$0.134	$5.49	Not stated
Modal	Per second of compute	Not applicable	About $3.95	$30 per month (Starter)

Frequently asked questions

How much did fal raise and what is it worth?

Fal announced a $1.3 billion Series F at a $16 billion valuation on November 3, 2026, according to TechCrunch. Bessemer Venture Partners led the round, with Nvidia, Sound Ventures and T. Rowe Price participating. The valuation is double the $8 billion it carried in April.

What does fal do?

Fal is a San Francisco company that hosts and runs AI video, image and audio models for developers, billing model APIs per output and serverless GPUs per second, according to TechCrunch and its pricing page.

Is fal the cheapest image and video API?

No. On the pricing pages we checked on November 4, 2026, Nano Banana 2 was $0.08 per image on fal against $0.05 on Together AI and $0.045 on WaveSpeedAI. Prices vary by model, resolution and settings, so compare the specific model you plan to use.

What are the best alternatives to fal?

Replicate for a large community model catalog, Runware and WaveSpeedAI for low published per-output prices, Together AI if you also run language models and dedicated GPUs, and Modal if you want to deploy your own code on serverless GPUs.

Who owns Replicate?

Cloudflare announced it would acquire Replicate in November 2025, and its FY2025 10-K records the acquisition. We could not confirm the exact closing date from the sources we checked.

Which platform is best for custom models?

Modal, because you ship your own code and weights and pay per second of compute. Fal also offers serverless GPUs, with H100 at $4.50 per hour list and discounted rates as low as $2.49.

How we evaluated

We confirmed the funding news from TechCrunch's report and a second independent search with consistent coverage. We picked platforms that serve image and video models through an API or sell serverless GPUs for running them. We read each vendor's public pricing page on November 4, 2026 and report only numbers that appeared there. Listed prices are usually minimums at the lowest resolution, duration or step count, so real costs run higher and the same model is not always the same configuration across platforms. We did not benchmark latency, quality or throughput. Baseten and Fireworks AI, which TechCrunch also names as fal competitors, are excluded because their public pages did not list comparable image or video prices when we checked. Ranking weights catalog breadth, price transparency and billing fairness.

Catalog breadth for current image and video models
Published, checkable pricing and what is not billed
GPU access for custom models
Free credit and onboarding friction

Innvesti may have commercial relationships with some companies mentioned. Our analysis is editorially independent. Figures cited are sourced; where reliable data is unavailable we say so rather than estimate.

Key takeaways

What the round does and does not tell you

The same model, different prices

Hosted catalog or your own code

Billing details that change real cost

1. fal

Pros

Cons

2. Replicate

Pros

Cons

3. Runware

Pros

Cons

4. WaveSpeedAI

Pros

Cons

5. Together AI

Pros

Cons

6. Modal

Pros

Cons

Frequently asked questions

How we evaluated

Sources