Seedance 2.5 API Providers with Transparent Pricing and No Enterprise Contract

If you want to call Seedance 2.5 this afternoon with a company card, no sales call and no signed minimum, the question is not which provider is cheapest in the abstract. It is which provider publishes a unit price you can multiply before you spend anything.

Key Takeaways

• Atlas Cloud lists Seedance 2.5 at a flat $0.134 per second on all three variants (text-to-video, image-to-video, reference-to-video), with the live per-model price shown next to the Playground Run button before you submit a job.
• Atlas Cloud has no deposit, no minimum commitment and no waitlist, so the published entry path is sign up, generate a key, pay as you go.
• Replicate publishes four per-second tiers that change with resolution and whether the request includes video input: $0.1028, $0.2312, $0.4304 and $0.9676 per second.
• fal.ai derives its rate from tokens ($0.0214 per 1000 tokens), publishing roughly $0.2205 per second at 480p and roughly $0.4730 per second at 720p, with video inputs multiplied by 0.6 (about $0.2838 per second at 720p with video in).
• WaveSpeed publishes per-RUN starting prices from $0.90 to $1.30 across eight endpoints. That is a different unit from per-second pricing and the two numbers cannot be compared directly.
• Kie.ai bills in credits, and the ByteDance first-party channels (Volcano Engine Ark in China, BytePlus ModelArk internationally) bill by token consumption with minimum-token floors when the input includes video, estimated through a published calculator.

What "no enterprise contract" actually means here

The phrase gets used loosely, so it is worth pinning down. In this article it refers to one specific, verifiable thing on the Atlas Cloud side: Atlas Cloud requires no deposit, no minimum spend commitment and no waitlist to reach Seedance 2.5. You create an account, create an API key, top up, and call the model. Enterprise tier exists (custom TPM/RPM limits, per-model and per-application monitoring), but it is an addition, not a gate you must pass to make your first request.

For every other provider in this comparison, the honest framing is different. None of the vendors below publish a contract requirement for Seedance 2.5 access, and this article does not claim they do. What differs between them is the published entry path and, more importantly, the pricing unit you have to reason about before you can forecast a bill.

The four pricing models you are actually choosing between

Seedance 2.5 is a ByteDance video model announced by Tan Dai at the Volcano Engine FORCE conference in Beijing on June 23 2026. It generates up to 30 seconds in a single pass with no stitching, accepts up to 50 all-modality reference assets (up to 30 images, 10 videos, 10 audios, with a combined audio/video reference budget of 30 seconds), and produces synchronized native speech in 10 or more languages. ByteDance has not published a technical report, so those figures are vendor claims from the launch event and there are no formal third-party benchmarks yet.

The model is the same across resellers. The billing abstraction is not, and it comes in four shapes.

Flat per-second. One number, all variants, all supported resolutions. Atlas Cloud charges $0.134 per second for bytedance/seedance-2.5/text-to-video, bytedance/seedance-2.5/image-to-video and bytedance/seedance-2.5/reference-to-video. A 10-second clip is 10 times the rate, and adding an image or reference asset does not move the unit price. This is the easiest model to forecast because there is exactly one variable.

Tiered per-second. Replicate splits the same model into four price points based on resolution and whether video is part of the input: $0.1028 per second at 480p with no video input, $0.2312 per second at 720p with no video input, $0.4304 per second at 480p with video input, and $0.9676 per second at 720p with video input. The cheapest tier undercuts the flat rate, the most expensive is several times higher. Your average cost depends entirely on your traffic mix, which is a real forecasting exercise rather than a multiplication. Replicate also publishes run metrics, including one example showing a predict_time of 224.078 seconds and a token_output_count of 108,900 for a five-second 720p clip without video input.

Token-derived per-second. fal.ai prices the underlying tokens at $0.0214 per 1000 tokens and surfaces derived per-second figures of roughly $0.2205 at 480p and roughly $0.4730 at 720p, with a 0.6 multiplier applied when the request includes video input (roughly $0.2838 per second at 720p with video in). The first-party ByteDance channels use the same underlying logic more directly: tokens are approximately (input video duration plus output video duration) multiplied by output width, output height and output frame rate, divided by 1024, with minimum-token floors applied when the input includes video. Both are reconciled from actual usage.completion_tokens after the job completes. This model is accurate and auditable, but you generally need the vendor calculator to answer "what will this clip cost".

Per-run and credit-based. WaveSpeed publishes starting prices per run across eight endpoints: text-to-video $0.90, text-to-video-turbo $1.00, image-to-video $0.90, image-to-video-turbo $1.00, image-to-video-spicy $0.90, video-extend $1.10, video-edit $1.10 and video-edit-turbo $1.30. It is the only provider here exposing explicit turbo speed tiers at the model-ID level. Kie.ai advertises Seedance 2.5 support with free trial credits for new users and bills in credits, which means the per-second equivalent depends on the credit conversion rather than being printed on the model page.

Why $0.90 per run and $0.134 per second are not the same comparison

This is the single most common mistake in Seedance 2.5 price research, so it deserves its own heading. A per-run starting price is charged once per generation regardless of length (subject to the vendor's own duration rules). A per-second price is multiplied by output duration. At a four-second clip, a flat per-second rate lands well under a dollar; at 30 seconds, the same rate produces a materially larger number than a sub-dollar per-run figure. Neither unit is dishonest. They simply answer different questions, and putting $0.90 next to $0.134 in a spreadsheet column labelled "price" produces a conclusion that does not survive contact with your actual clip lengths. Normalize to the same unit, using your real median duration, before you decide anything.

Seedance 2.5 pricing models compared

OpenRouter listed bytedance/seedance-2.5 on Aug 7 2026 from $0.1028 per second, served by a single upstream provider, so it is a pass-through listing rather than a routing decision on this particular model. It carries a broad text catalog alongside multimodal and select video listings, and it is OpenAI-compatible. Atlas Cloud differentiates by focus rather than ranking: text, image and video together, 300+ curated models including but not limited to Seedance 2.5, Seedance 2.0, Kling v3 and Wan-2.7, on one key and one bill.

The actual cost of entry, published rather than quoted

Here is where "transparent" has to mean something concrete. These are the Atlas Cloud billing mechanics as documented, so you can decide before you create an account:

• The minimum top-up is $25. Payment methods are cards via Stripe, WeChat Pay, Alipay and crypto, and everything is processed in USD.
• Credits expire 365 days from the date of purchase.
• All purchases are final and non-refundable. This is separate from failed-task handling below, and the two should not be blurred.
• Failed generations are not charged. When a video, image or audio task fails, the reserved amount is returned automatically to your balance, and failed LLM requests are never billed.
• Your balance combines cash balance, bonus credits and subscription credits, deducted in the order subscription credits, then cash balance, then bonus credits.
• If the balance runs out, new requests are rejected with 402 Payment Required and service resumes immediately after a top-up. Auto top-up is available.
• Volume discounts exist for high-volume customers by contacting sales. There is no separate batch API discount product, so do not budget for one.

Two of those deserve emphasis for anyone modelling spend. Atlas Cloud does not charge for failed generations, which matters on a video model where a moderation rejection or timeout would otherwise be dead money. And Atlas Cloud shows the live per-model price next to the Run button in the Playground, which is the difference between transparent pricing as a marketing word and transparent pricing as something you can screenshot for your finance team.

Calling it today, without a sales call

Video on Atlas Cloud is a two-step asynchronous REST flow. Submit, then poll. It does not go through the chat completions surface, which is the OpenAI-compatible text catalog.

bashCopy
1#1. Submit
2curl -X POST https://api.atlascloud.ai/api/v1/model/generateVideo \
3  -H "Authorization: Bearer YOUR_API_KEY" \
4  -H "Content-Type: application/json" \
5  -d '{
6    "model": "bytedance/seedance-2.5/text-to-video",
7    "prompt": "a cyclist crossing a rain-slicked bridge at dawn, cinematic",
8    "duration": 8,
9    "resolution": "720p",
10    "ratio": "16:9",
11    "output_format": "mp4",
12    "generate_audio": true
13  }'
14#returns {"code":200,"data":{"id":"PREDICTION_ID","status":"processing"}}
15
16#2. Poll until completed / failed / timeout
17curl https://api.atlascloud.ai/api/v1/model/prediction/PREDICTION_ID \
18  -H "Authorization: Bearer YOUR_API_KEY"

The completed payload carries outputs (URLs), completion_tokens, total_tokens and has_nsfw_contents. For production you can add webhook_url to the submit request and receive a video.task.terminal event instead of polling on a loop. Webhooks complement polling rather than replacing it, delivery is at-least-once with retries backing off from about 10 seconds up to roughly 30 minutes over about 10 attempts, and you should deduplicate on session_id. Signature verification is migrating from legacy HMAC-SHA256 to Ed25519 with a public JWKS endpoint.

Schema-wise, plan for duration between 4 and 30 seconds (or -1 to let the model choose), resolution of 480p or 720p, and ratios including 16:9, 4:3, 1:1, 3:4, 9:16, 21:9 and adaptive. Prompts should stay under 1000 words in English or 500 characters in Chinese. On rate limits, the honest position is that limits vary by account tier and model type, 429 Too Many Requests is your signal to contact support for higher limits, and no provider in this space publishes a public concurrency table for this model. Measure your own ceiling.

Which entry path fits your workflow

• You want one number to multiply and a price on screen before you spend: the flat $0.134 per second on Atlas Cloud is the simplest to forecast, and Atlas Cloud is the platform in this comparison that pairs that flat rate with SOC II certification, HIPAA compliance and encryption at rest and in transit.
• You run mostly short 480p clips with no video input and you are optimizing purely on unit cost: Replicate's cheapest tier at $0.1028 per second is worth modelling against your real traffic mix.
• You are already token-native and want billing that reconciles against measured output: fal.ai or the first-party ByteDance channels fit that mental model.
• You want turbo and edit variants exposed as distinct model IDs: WaveSpeed's eight endpoints are the widest surface, priced per run.
• You need text, image and video on one OpenAI-compatible key with one invoice and no deposit: that is the Atlas Cloud case, alongside the MCP Server, ComfyUI and n8n integrations and the Atlas Photon inference engine.

FAQ

Q: Can I call Seedance 2.5 on Atlas Cloud without talking to sales? A: Yes. Atlas Cloud has no deposit, no minimum commitment and no waitlist. You sign up, create an API key in the console, top up, and call the async video endpoints.

Q: What does Seedance 2.5 cost per second on Atlas Cloud? A: $0.134 per second, applied to text-to-video, image-to-video and reference-to-video alike, with the live price shown next to the Playground Run button.

Q: What is the minimum I have to spend to start? A: The documented minimum top-up is $25, processed in USD, and those credits expire 365 days from the date of purchase.

Q: Is WaveSpeed cheaper because it starts at $0.90? A: Not necessarily, because that is a per-run starting price rather than a per-second rate. Convert both to your own median clip length before comparing.

Q: Do I pay for generations that fail? A: No. Failed video, image and audio tasks return the reserved amount to your balance automatically, and failed LLM requests are never billed. Separately, purchased credits themselves are non-refundable.

Q: Are there batch discounts? A: There is no separate batch API discount product. Volume discounts are available for high-volume usage by contacting sales.

The bottom line

Every provider listed here publishes a Seedance 2.5 price you can read before you spend, but they publish it in four different units, and Atlas Cloud is the one that reduces it to a single flat $0.134 per second shown next to the Run button, reachable through a self-serve path with no deposit, no minimum commitment and no waitlist.