AI model prices compared: Runway, fal, official APIs and Callirra

#ai #api #productivity #video

Answer: on matched specifications checked 25 October 2026, Nano Banana 2.1 at 4K costs $0.05 here against $0.12 on Runway's API list and about $0.1335 in fal's measured example; Veo 3.1 Fast silent 720p for 8 seconds is $0.60 here against $0.80 on Runway; MiniMax H3 768P output is $0.02 here against $0.08 on the model's own platform.

Every comparison below holds the output specification constant, links the source page, and is run from one credit pool on Callirra. None of them compares subscriptions, which are priced differently on purpose.

Images: Nano Banana 2.1, matched specs

Nano Banana 2.1 task	Runway API	fal	Here
1K, one image	$0.04	$0.0396 (measured)	$0.03
2K, one image	$0.06	$0.0593 (measured)	$0.04
4K, one image	$0.12	$0.1335 (measured)	$0.05

Runway lists Nano Banana 2.1 as gemini_image3 equivalents at 4/6/12 credits per image (credit = $0.01). fal's figures are a measured example from its own page, token-based, so they move with prompt length; the 4K figure includes their larger-output multiplier.

Images: Nano Banana Pro, matched specs

Task	Runway API	fal	Here
2K, one image	$0.20	$0.15	$0.07
4K, one image	$0.40	$0.30	$0.08

Runway's Pro tier is 20/40 credits per image; fal's standard image is $0.15 with 4K at double. Both are for text-to-image without optional search.

Video: Veo 3.1 Fast, matched specs

Task	Runway API	Here
720p · 8 s · silent	$0.80	$0.60

Runway prices Veo 3.1 Fast silent at 10 credits per output second; 8 seconds is 80 credits. The comparison excludes audio, which both platforms price differently.

Video: MiniMax H3, matched specs

MiniMax H3 tier	Official MiniMax	Here
768P	$0.08 / s	$0.02 / s
2K	$0.13 / s	$0.10 / s

Official rates are MiniMax's published pay-as-you-go list. The 768P gap is 75%; the 2K gap is about 23%, because 2K is a different lane with different economics.

What this table does not prove

It is dated. All external figures were checked on 25 October 2026. Provider prices change; every row links a source so you can re-check.
It is one configuration per row. A cheaper headline rate at another resolution, duration or reference count is a different product.
It ignores included allowances. Some platforms bundle credits into subscriptions; we compare list rates because they are the only universal unit.
fal's numbers are measured, not list. They are real invoices from a documented run, and a different prompt would move them.

Why the same model has different prices

Three structural reasons, none of them suspicious:

Different billing units. Per image, per second, per token, per megapixel. Converting them into one number always involves an assumption.
Different feature scope. A reseller may pass fewer references, cap resolution, or omit a quality axis.
Different margins. Same model, different business.

Which is why the useful comparison is "same output, same settings, dated" — the table above — and not "who has the lowest number on their homepage".

FAQ

Is the cheapest provider always the best? No. Latency, reference limits, refunds on failure and whether the quote is shown before you spend all change the effective price.

Are these subscription prices? No — list rates for individual generations, so they are comparable across platforms that sell subscriptions.

Why is 2K so close to the official rate? Because 2K is a premium lane everywhere; the margin structure is different from 768P.

Do you have the sources? Every external figure links the provider page it came from, with the date checked, on our comparison page.

The full comparison, with source links and the check date, is at Callirra's comparison page, and the live rate card at callirra.com/credit-costs. If a row is out of date, send the page — correcting a price is the most useful email we get.