Wink Pings

One Key, Seedance at List Price: Sunra Lays Out the API Bill

Sunra launched an aggregated API: one Key to call image, video, audio, and 3D models, paying at each vendor’s own API price, pay as you go, with no subscription and no markup. The price list is out, and Tjay’s real-world test is a reminder: a low unit price does not mean low total cost—queues, failures, and retries all come back onto the bill.

Sunra has launched an API. The official wording is direct: one Key to call mainstream image, video, audio, and 3D models, with each model charged at its vendor’s own API price, pay as you go, no subscription, no markup. Enterprise buyers can get up to 40% off the list price of leading models. Source: Sunra official tweet.

Sunra tweet image

The price list is there too:

  • Seedance 2.5: $0.231/second, 720p, audio on
  • Kling 4.0: $0.157/second, 720p, audio on
  • GPT Image 2.5: $0.0527/image, 1024×1024, high

Sunra price list

When AmirMušić retweeted it, he wrote, “Seedance 2.5 follows ByteDance’s own pricing,” followed by “Let the tax free generation begin.” The point people in the comments fixated on is practical: Seedance 2.5 and Kling 4.0 sit behind the same Key, with no markup, and they’ll test it tonight. Others said they can finally stop switching among six API dashboards. That picture is not exaggerated. When one project needs image, video, and audio at the same time, developers commonly end up topping up several dashboards separately, checking quotas separately, and reconciling the bills at the end.

The official post added a couple of product claims: direct upstream partnerships with model vendors mean fast, secure, reliable requests; whichever model you call is the one that runs; and every request and charge is itemized in the console. For people building products, the latter is more useful than “fast.” Only when the bill can be broken down do you know whether cost went to which model, which retry, or which resolution.

But the price list is only the first layer. Tjay posted a cost-optimization guide for video models, claiming to spend $70K a month on tokens. He tested Higgsfield, Runware, KIE, Fal, BytePlus, Runway, Sume, Sunra, and Replicate. The conclusions are not uniform: the cheapest option depends on which model you need to use. KIE is cheap on Google models; he even suspects it buys $200 retail accounts and then uses browser automation to provide the service. BytePlus’s own Seedance and Seedream are cheapest on BytePlus, and only it supports Digital Asset Library, which costs $1,400 a month. Higgsfield Soul 2.0 is only available on Higgsfield and is also cheapest there. Sume he called misleading marketing because its comparison used lower-resolution generations.

Reliability is another ledger. Tjay says Fal is the most stable; testing a 1990s Pixar-animation-style Seedance 2.5 video took about 17 minutes end-to-end. The same task on other providers sometimes takes 45 minutes, mostly spent queuing, with insufficient capacity and stuck requests. Runware and Sume have more obvious dropped-request problems; some requests stay processing forever. He also mentions that output quality differs across providers: Fal is best, Runware and Replicate are slightly worse, with signs of resolution being downgraded. These are Tjay’s tests and judgments, not official specs.

His money-saving tips are very specific: generate at 480p first, then upscale with Topaz, and cost can drop to one quarter; for portrait workflows use Wan 3.0, which is 64% cheaper than Seedance 2.5 with comparable quality; for animation workflows use Kling 4, which is 34% cheaper with comparable quality; reserve Seedance 2.5 only for talking heads. Someone in the comments added a sharp point: calculate cost per usable clip, not per generation. A model can be 3x cheaper but require 4 retries before it produces something usable, making it more expensive in the end.

Tjay’s BytePlus spending proof

Amir reply image

Amir reply image

Sunra removes the “markup layer,” and that is where the value lies: the barrier to trial and error is lower; indie developers and content teams no longer have to be locked in by subscription fees first; they can pay per request and run a small workflow to see the result. But “list price” does not equal “lowest total cost.” Queue time, failure rate, retry count, post-processing, and asset-library monthly fees all come back onto the bill. Competition among aggregated APIs will also shift from “who has the lowest unit price” to “who reliably delivers usable output and who itemizes the bill clearly.”

If you want to try it, don’t just look at the price list first. Run it on your real workflow and calculate cost based on the clips you can actually use. Price lists look best; bills are the most honest.

发布时间: 2026-10-29 20:37