fal.ai Pricing, November 2026: What Each Model Costs

Checked November 2, 2026. fal has no plan to pick: you buy prepaid credits and every model has its own price. For a standard 1024×1024 image, that price runs from $0.003 (FLUX.1 [schnell]) to about $0.21 (GPT Image 2 at its default high quality). A 5-second 720p video costs from $0.05 (Boreal) to $2.37 (Seedance 2.5). Speech costs $0.004 to $0.10 per 1,000 characters.

Three things are worth knowing before you read the tables:

• GPT Image 2.5 costs a quarter of GPT Image 2 at default settings: about $0.053 against $0.211 per 1024×1024 image at high quality.
• Veo 3.1 leaves fal on November 17, 2026. The Veo 3.1, Fast and Lite pages all carry that deprecation notice, and Veo is already gone from fal's public catalog.
• Two video discounts run out this month. MiniMax H3 Max is 30% off until November 15, and FLUX.3 Max video is 75% off until November 7.

This page lists one vendor's prices. To compare fal with Replicate, OpenAI, Google and BytePlus, see our AI image and video API pricing comparison.

How fal bills

fal's pricing docs describe a prepaid credit balance that each request draws down. Models bill by output, and the unit depends on the model. We read the billing unit from the page of all 989 image, video, speech and music endpoints in fal's catalog:

As of 2026-11-02 from fal's model catalog and the billing data on each model page. How we computed it: we pulled all 1,464 catalog entries, kept the seven generation categories (text or image to image, text or image to video, text to speech, text to audio, speech to text), and grouped the billing unit each page reports.

The base price in fal's billing data is not always what you pay. For many models it is a placeholder. Grok Imagine Video 1.5, for example, lists a base of 0.01 per second, while its page charges $0.14 per second at 720p. Every number below comes from the model page's own pricing statement or table, not from the base price.

Other rules from fal's FAQ:

• Server errors (HTTP 500+) and queue time are free. Invalid inputs (HTTP 422) can still be charged if a GPU already worked on the request.
• New accounts can run 2 requests at once. The limit rises automatically with credit purchases, up to 40.
• Purchased credits expire after 365 days. Free credits and coupons last one week to one year.

Image models: cost per image

As of 2026-11-02 from each model's fal page. How we computed it: one text-to-image output at 1024×1024 (1 megapixel), default settings unless the row names a quality level, no reference images, no web search. For token-billed models we used the per-size table on the fal page. fal calls Seedream 5.0 Pro's price tentative. GPT Image rows exclude prompt tokens, which fal says add about $0.0001 for a short prompt. "About" marks fal's own estimates: Nano Banana 2.1 assumes a short prompt with medium thinking, and the MAI figures are fal's averages.

What changes the price:

• Resolution. Per-megapixel models scale with area: FLUX.3 Klein is $0.009 at 1K and $0.036 at 4K. Nano Banana 2 charges 1.5× for 2K and 2× for 4K. Nano Banana 2.1 is about $0.059 at 2K and $0.134 at 4K.
• Quality settings on GPT Image. fal defaults GPT Image to high quality. GPT Image 2.5 at 1024×1024 is $0.006 low, $0.013 medium, $0.053 high, $0.094 xhigh and $0.211 max. The Flare and Sunburst variants show the same table. If you leave the default on GPT Image 2, you pay $0.211, 35 times its low-quality price.
• Inputs count on some models. Flux 2 Flex bills $0.05 per megapixel on both input and output, so editing with a reference image costs more than generating. Krea 3 also counts each reference image.
• Web search adds $0.015 per image on Nano Banana 2 and Nano Banana Pro.

For which of these models produce the best images, see our best AI image models guide.

Video models: cost of a 5-second clip

As of 2026-11-02 from each model's fal page. How we computed it: 5 seconds of text-to-video at 720p, or 768p where the model has no 720p tier. Standard prices first, with active promotional prices and their end dates in parentheses. "With audio" splits are the page's own rates. Seedance is billed in tokens, so its per-second figures are fal's own approximations. "Included" means the page or model card says sound is generated with the video. "Not stated" means the pricing statement doesn't mention audio. It does not mean the model is silent.

What changes the price:

• Resolution. 1080p usually costs 1.5 to 2.5 times the 720p rate. Kling 4 is $0.235 per second at 1080p, Wan 3.0 is $0.20 and Seedance 2.0 is about $0.68. Kling 3 at 4K is a flat $0.42 per second.
• Audio. Kling 3 Pro adds 50% for sound ($0.112 to $0.168 per second), and $0.196 with voice control. Kling 2.6 Pro doubles.
• US-hosted Seedance costs more. Seedance 2.0's US endpoint charges about $0.37 per second at 720p, against $0.303 on the standard one.
• Image-to-video costs about the same. Kling 4 and Seedance 2.5 charge their text-to-video rates for image-to-video.

Veo 3.1 users: move before November 17. fal's Veo 3.1 pages show $0.20 per second without audio and $0.40 with audio at 720p or 1080p, plus the deprecation date. Veo is no longer listed in the catalog, so new users won't find it. Our best AI video models guide ranks the alternatives by output, not price.

Voice, transcription and music

As of 2026-11-02 from each model's fal page and fal's pricing page. HeyGen's two voice models were on a 50% promotion until October 31. The rates above are the regular prices the pages list.

Transcription and music on fal, same date:

• ElevenLabs Scribe v2: $0.008 per minute of input audio, or $0.48 per hour. Keyterm prompting adds 30%.
• Kotoba STT: $0.25 per hour of input audio.
• ElevenLabs Music v2 and v2.5: $0.60 per minute of output, rounded up, so a 30-second clip costs the same as a full minute.
• MiniMax Music 2.6: $0.15 per generation.
• Sonilo 1.1: $0.0025 per second of output, per sample.

For quality rankings, see our best AI voice models and best AI music generators guides.

Running your own model: GPU prices

If a model isn't in the catalog, you can deploy it on fal's serverless GPUs, billed per second while your runners are alive. fal's pricing page lists an H100 (80 GB) at $4.50 an hour, "as low as" $2.49. An H200 is $6.00 ("as low as" $2.99), a B200 is $7.99 ($5.49) and a B300 is $12.99 ($5.99). The page doesn't say what qualifies for the lower price. Ask sales before you plan around it.

How to keep a fal bill down

1. Set quality explicitly on GPT Image. fal's default is high. Low quality costs about $0.006 a frame, which is enough for drafts and thumbnails.
2. Draft small, then render once. Video at 480p costs roughly half to two-thirds of the 720p price on the models in the table that list both. Lock the prompt before paying for 1080p.
3. Turn off audio you'll replace. On Kling 3 Pro, that saves a third of the clip price.
4. Read the unit before you batch. A strictly per-megapixel model like Z-Image Turbo costs four times as much at 2048×2048 as at 1024×1024. A flat per-image model like Seedream 5.0 Lite costs the same at both.
5. Check promo end dates. MiniMax H3 Max's 30% discount ends November 15, and standard rates start November 16.

Using fal through Teamday

Teamday's Images and Video modules can run four fal models: Flux 2 Flex for images, and Kling 3 Pro, Kling 2.6 Pro and Wan 2.6 for video. Connect your fal API key and those runs bill to your fal account at the prices above. The same modules also run Seedream, Seedance, GPT Image, Gemini, Wan, Qwen Image and Grok Imagine directly from their makers.

Related

• AI image and video API pricing: fal vs Replicate vs OpenAI vs Google
• FAL.AI vs Replicate
• Best AI image models
• Best AI video models