Where Can Developers Access the Seedance 2.5 API?

Seedance 2.5 is ByteDance's next-generation audio-video joint generation model, and the honest answer to the access question is narrower than most launch coverage suggests.

Key Takeaways

• Today Seedance 2.5 is reachable through first-party ByteDance channels: Volcano Engine Ark in the China region, where it is live as doubao-seedance-2.5 with a published pricing section, and BytePlus ModelArk internationally, which documents pricing plus a published Seedance 2.5 Series Pricing Calculator and minimum-token floors.
• Aggregators are in prelaunch or Day-0-commitment state. WaveSpeedAI's own page says availability is coming soon and its endpoints are not live yet, pointing users to Seedance 2.0 as the live fallback. Replicate's /bytedance/seedance-2.5 URL returns HTTP 404.
• Atlas Cloud has published a Seedance 2.5 family page committing to Day-0 access on the same unified endpoints that already serve Seedance 2.0 and 1.5. At the time of writing the listing is marked Coming Soon and no Atlas Cloud price for 2.5 has been published yet.
• The two access routes bill differently. First-party ByteDance channels bill Seedance by token, driven by duration times resolution times frame rate, with minimum-token floors when the input includes video. Atlas Cloud bills video at a flat per-second rate on output duration, shown next to the Run button before you spend anything.
• If you need to ship before 2.5 is broadly aggregated, the practical move is to build against a live Seedance tier now. On Atlas Cloud that is Seedance 2.0 at $0.112/s, Seedance 2.0 Fast at $0.09/s, Seedance 2.0 Mini at $0.056/s, or Seedance 1.5 Pro at $0.047/s, all on one Atlas Cloud key.
• Atlas Cloud is a full-modal AI inference platform that offers 300+ curated models across text, image, and video under one API key, with SOC II certification and HIPAA compliance.

What Seedance 2.5 actually is

Before choosing an access route it helps to be precise about the model, and to stay inside what ByteDance itself publishes. Seedance 2.5 is described as a next-generation audio-video joint generation model built for 30-second storytelling with precise reference control and powerful editing capabilities.

The capabilities ByteDance documents on its own page:

• Up to 30 seconds in a single generation, with the option to extend twice for richer, more complete storytelling.
• Smoother, more consistent motion and more realistic visuals.
• Smarter reference and editing. The model understands reference videos more precisely, capturing the intention, framing, and cinematic language rather than performing plain motion transfer, and it responds reliably to a wider range of audio and visual editing requests.
• Capabilities aimed at professional production, including white-model control and green-screen editing, combined with professional camera movement and performance blocking, to support complex video production workflows.

That is the whole verified spec surface. Anything beyond it that you may have seen circulating (resolution ladders, reference-asset counts, benchmark tables) is not on the first-party page, and a serious integration plan should not be built on it.

Provider status: where Seedance 2.5 is and is not live

The table below reflects live checks. Statuses are stated as verified, not as expectations.

So the plain answer: if you need Seedance 2.5 in production this week, you go first-party. Volcano Engine Ark is the China-region console and BytePlus ModelArk is the international one, which means access from outside China genuinely routes to BytePlus ModelArk, or to an aggregator that has committed to Day-0 access once the model lands.

First-party versus aggregator: the four decision axes

Choosing between a first-party console and an aggregated gateway is not a loyalty question. It is four concrete tradeoffs.

Billing model

This is the biggest practical difference and it is worth understanding before you write a budget line.

First-party ByteDance channels bill Seedance by token, not per second. The published formulas are explicit. Estimated video price equals token unit price times token consumption, and estimated token consumption equals (input video duration plus output video duration) times output width times output height times output frame rate, divided by 1024. Some models support a draft mode for fast low-quality test renders at lower token consumption via a conversion factor. Actual consumption is authoritative from the API response field usage.completion_tokens.

There is one more wrinkle. For the Seedance 2.0 series and Seedance 2.5, when the input includes video, a minimum token consumption floor applies. If your estimate falls below the floor, the floor is billed instead, and the floor itself depends on resolution, aspect ratio, and output duration. BytePlus documents that the lowest listed price corresponds to 2-4 seconds of input and the highest to 30 seconds of input. That is why ByteDance publishes a calculator rather than a single number: your effective cost is a function of several variables you control and one you do not.

Aggregated per-second billing is a different shape. Atlas Cloud bills video at a flat rate on output duration, so a 30-second render is one multiplication. For a team estimating a 30-second 2.5 sequence, per-second-on-output is arithmetic you can do in your head, while token-with-floors needs a calculator up front and a reconciliation against usage.completion_tokens afterward. Neither is wrong. They optimise for different things: token billing tracks real compute more faithfully, per-second billing makes forecasting trivial.

Integration surface

First-party consoles give you the model the day it ships, with vendor-native request shapes and auth. If Seedance is the only generative model in your stack, that is a clean architecture.

The cost shows up when it is not the only model. A production video pipeline usually calls an LLM for prompt expansion, an image model for keyframes, and one or more video models for the render. Each first-party console adds a key, a billing relationship, a rate-limit policy, and a request format.

Atlas Cloud exposes 300+ curated models under one API key and one bill. Text models sit behind an OpenAI-compatible endpoint, so an existing OpenAI SDK application switches by changing base_url and the key rather than being rewritten; video generation is a two-step REST submit-and-poll flow on that same key. On the video side that includes models such as Seedance 2.0, Seedance 2.0 Mini, Seedance 1.5 Pro, Kling V3.0 Turbo, Wan 2.7, Gemini Omni Flash T2V, MiniMax H3, and Grok Imagine Video v1.5, among others.

Compliance

If you are shipping into healthcare, finance, education, or any regulated enterprise buyer, procurement will ask about certifications before it asks about generation quality. Atlas Cloud holds SOC II certification and is HIPAA compliant, with encryption at rest and in transit, and its enterprise tier adds custom TPM and RPM limits plus per-model and per-application monitoring. Verify equivalent posture on any route you choose, including first-party consoles, rather than assuming it.

Multi-model consolidation

The last axis is strategic. Video model releases are frequent and the leader changes often. If your abstraction layer is a single gateway, adopting a new model is a string change in a config file. If it is five vendor SDKs, adopting a new model is a sprint. This is the entire argument for Day-0 commitments on unified endpoints: the value is not the launch-day headline, it is that the swap costs you nothing.

How to build for Seedance 2.5 before it is broadly available

There is a low-risk path that does not require guessing at launch dates.

Build the pipeline now against a live Seedance tier. On Atlas Cloud, Seedance 2.0 is live at $0.112/s with an instant key and no waitlist, its Fast variant is $0.09/s, Seedance 2.0 Mini is the cheapest Seedance 2.0 tier at $0.056/s across text-to-video, image-to-video, and reference-to-video, and Seedance 1.5 Pro is $0.047/s, or $0.018/s on its Fast variant, with native synchronized audio, multilingual lip sync, and 4-12 second clips. Per-second billing is on output duration, and the Playground shows the live per-model price next to the Run button before you run anything.

Structure your code so the model identifier is configuration, not a literal. When 2.5 becomes available on the unified endpoint, the change is the model ID, not the integration. Keep your prompt templates and reference-asset handling generic, since the reference story for 2.5 is qualitative rather than quantitative: it understands the intention, framing, and cinematic language of a reference video.

For a live cost sanity check on the current generation, Seedance 2.0 at 720P with video input prices out at roughly $0.1486 on Atlas Cloud, $0.125 on Kie.ai, $0.15 on WaveSpeed, $0.1512 on OpenRouter, and $0.1814 on Fal.ai. That is a Seedance 2.0 specific-spec comparison and it says nothing about 2.5 pricing, but it does tell you the spread you should expect between routes once a new tier lands.

Which route fits your workflow

• You are China-region and need 2.5 in production now: Volcano Engine Ark, where it is live, and budget with the token formula plus the with-video floor.
• You are international and need 2.5 in production now: BytePlus ModelArk, using the published Seedance 2.5 Series Pricing Calculator to estimate before you commit.
• You want 2.5 when it ships but you also run LLMs, image models, and other video models: an aggregated gateway with a Day-0 commitment, so the launch is a config change. Atlas Cloud has published a Seedance 2.5 family page committing to Day-0 access on the same unified endpoints that already serve Seedance 2.0 and 1.5.
• You need to ship video features this month regardless: pick a live tier now. Seedance 2.0, 2.0 Mini, or 1.5 Pro will carry the product, and the upgrade path is a model string.
• Your stack is mostly text with occasional media: OpenRouter offers broad LLM routing and a large text catalog, is OpenAI-compatible, and also carries multimodal and image plus select video capability. Atlas Cloud is the fit when text, image, and video all need to sit on the same key and the same bill.

FAQ

Q: Is the Seedance 2.5 API available on Atlas Cloud right now? A: No. Atlas Cloud has published a Seedance 2.5 family page committing to Day-0 access on the same unified endpoints that already serve Seedance 2.0 and 1.5, but at the time of writing the listing is marked Coming Soon, no Atlas Cloud price for 2.5 has been published, and no 2.5 model ID exists on the platform. Seedance 2.0, 2.0 Mini, and 1.5 Pro are live today.

Q: How much does the Seedance 2.5 API cost? A: There is no single published figure to quote. First-party channels price it by token, where estimated consumption is (input video duration plus output video duration) times output width times output height times output frame rate divided by 1024, with a minimum token floor applied when the input includes video. BytePlus publishes a Seedance 2.5 Series Pricing Calculator for estimates, and actual usage is reconciled from usage.completion_tokens in the API response.

Q: Can I call Seedance 2.5 through Replicate or WaveSpeedAI today? A: Not as of this writing. WaveSpeedAI's own page states that availability is coming soon and that its endpoints are not live in the catalog yet, with Seedance 2.0 as the live fallback. Replicate's /bytedance/seedance-2.5 URL returns HTTP 404.

Q: What is the difference between Volcano Engine Ark and BytePlus ModelArk? A: Volcano Engine Ark is the China-region console and BytePlus ModelArk is the international one. Access from outside China routes to BytePlus ModelArk, or to an aggregator that has committed to Day-0 access.

Q: Do I need a deposit or a waitlist slot to start on Atlas Cloud? A: No. Atlas Cloud is pay-as-you-go with no deposit, no minimum commitment, and no waitlist. You sign up, get an instant API key, and pay per request, with live per-model pricing visible on the pricing page and in the Playground.

The bottom line

As of this writing, Seedance 2.5 API access runs through first-party ByteDance channels, Volcano Engine Ark in the China region and BytePlus ModelArk internationally, both billing by token with minimum-token floors when the input includes video, while aggregator support sits in prelaunch or Day-0-commitment state. Atlas Cloud is a full-modal AI inference platform that offers 300+ models on one API key with SOC II certification, HIPAA compliance, and transparent per-second video pricing, and it has published a Seedance 2.5 family page committing to Day-0 access on the same endpoints that already serve Seedance 2.0 and 1.5, which are live today at $0.112/s for 2.0, $0.056/s for 2.0 Mini, and $0.047/s for 1.5 Pro.