ByteDance API Pricing: Every Model, Every Tier (2026)

ByteDance API pricing is three units and one parameter

Search “ByteDance API pricing” and you get numbers that cannot be subtracted from each other: from $0.06726/second, $5.6 per million tokens, flat $0.09/s, $0.045 per image. They describe the same family of models — Seedance for video, Seedream for images — metered three different ways. ByteDance API pricing is a unit plus a parameter, not a price list. The unit tells you what you are buying (a second of output video, one generated image, or a slice of a token budget). The parameter tells you what changes the number: for every ByteDance video model on sale at Modellix’s ByteDance provider page as of September 13, 2026, that parameter is resolution — one switch, and nothing else; the Seedream image routes are flat per image.

The short version, all live-checked today: on the Modellix route, Seedance video runs from $0.0210/second (bytedance/seedance-2.0-mini-v2v, 480p) to $0.7800/second (bytedance/seedance-2.0-t2v and -i2v, 4K), and Seedream images are $0.0350/image (5.0 Lite) or $0.0810/image (5.0 Pro). At BytePlus/ModelArk — ByteDance’s own platform — the same families are billed per million tokens for video and by pixel tier for images. That difference is not a rounding detail; it decides whether two published prices can be compared at all.

Modellix distributes these models, so this breakdown comes from a seller of the routes it describes. Where the two platforms bill in the same unit, the lower number is not always ours — you will see both directions below.

What this breakdown covers (and what it leaves out)

Before the tables, the boundary. This page covers the media models ByteDance ships under the Seedance (video) and Seedream (image) names, priced from the live model pages on September 13, 2026.

• Included: all 17 ByteDance model pages currently on sale at Modellix — 12 Seedance video routes and 5 Seedream image routes. Model counts differ by definition, so be precise about which one you mean: 17 is the count for this provider’s on-sale pages today (the provider page’s own structured data declares numberOfItems: 17), not the platform-wide model-page count.
• Not included — text models: ByteDance also sells Seed LLMs (Seed 1.6, Seed 2.0 Lite and friends), which bill per million tokens. That is a different product line with its own catalog, priced per million tokens on ByteDance’s own platform — not on this page. Pages that mix Seed LLM token prices into a “ByteDance API pricing” answer are answering a different question.
• Not included — older generations: Seedance 1.x and Seedream 4.x are not on sale at Modellix as of today. Modellix carries Seedance 2.0/2.5 and Seedream 5.0. If you need a 4.x number, take it from ByteDance’s own platform — the BytePlus ModelArk pricing page lists seedream-4-0-250828 at $0.03/image (accessed September 13, 2026). We do not carry it, so we do not quote it as ours.
• Not included — free tiers: there is no signup credit to evaluate against. The $1 registration credit ended on August 19, 2026; trial quota is now requested by email (details in the access section below).

Every price in this article was pulled from the individual model pages on September 13, 2026, and each table row links to the page it came from. Prices on this platform move; re-check before you commit a budget.

All 17 ByteDance models on sale, and what each one costs

Five families, one page each. The unit column is the whole point: sec means the rate applies to every second of output video, img means a flat charge per output image.

Family	Models (provider page)	Unit	Range as listed
Seedance 2.5	seedance-2.5-t2v, -i2v, -v2v	per second	$0.0616 – $0.5690/sec
Seedance 2.0	seedance-2.0-t2v, -i2v, -v2v	per second	$0.0430 – $0.7800/sec
Seedance 2.0 Fast	seedance-2.0-fast-t2v, -i2v, -v2v	per second	$0.0330 – $0.1200/sec
Seedance 2.0 Mini	seedance-2.0-mini-t2v, -i2v, -v2v	per second	$0.0210 – $0.0800/sec
Seedream 5.0	seedream-5.0-lite, -lite-edit, -pro, -pro-edit, -pro-multi-reference	per image	$0.0350 – $0.0810/img

The provider page lists every on-sale ByteDance route with its price range in the card footer — the same ranges broken out by resolution below. Screenshot captured September 13, 2026 from www.modellix.ai/providers/bytedance.

Note what is not in that table: no duration surcharge, no audio surcharge, no “reference images cost extra” line. The ranges come from resolution alone, which is why the next two sections are the actual price list.

Seedance video pricing: per second, by resolution tier

Video is billed per second of output, and the rate depends on the resolution you request — not on the model’s generation time, not on how many reference images you pass. All numbers below are the per-second rates shown on each model page on September 13, 2026.

Model	480p	720p	1080p	4K
bytedance/seedance-2.0-mini-v2v	$0.0210	$0.0450	—	—
bytedance/seedance-2.0-fast-v2v	$0.0330	$0.0710	—	—
bytedance/seedance-2.0-mini-t2v	$0.0400	$0.0800	—	—
bytedance/seedance-2.0-mini-i2v	$0.0400	$0.0800	—	—
bytedance/seedance-2.0-v2v	$0.0430	$0.0930	$0.2280	$0.4670
bytedance/seedance-2.0-fast-t2v	$0.0600	$0.1200	—	—
bytedance/seedance-2.0-fast-i2v	$0.0600	$0.1200	—	—
bytedance/seedance-2.5-v2v	$0.0616	$0.1380	$0.3400	—
bytedance/seedance-2.0-t2v	$0.0700	$0.1500	$0.3700	$0.7800
bytedance/seedance-2.0-i2v	$0.0700	$0.1520	$0.3740	$0.7800
bytedance/seedance-2.5-t2v	$0.1030	$0.2310	$0.5690	—
bytedance/seedance-2.5-i2v	$0.1030	$0.2310	$0.5690	—

Source: Modellix model pages for each row, e.g. seedance-2.5-t2v, seedance-2.0-t2v, seedance-2.0-mini-v2v, all accessed September 13, 2026. Dashes mean the model does not offer that resolution — not that it is unpriced. None of the 12 video routes carried a model-level discount on this date.

Read the table by column, not by row. Three things fall out of it:

• The resolution spread is the budget risk, not the model choice. seedance-2.0-t2v is $0.0700/sec at 480p and $0.7800/sec at 4K — more than 11× on the same model, same prompt. Anyone budgeting from a “from $0.07/second” headline and then requesting 4K is off by an order of magnitude.
• The cheap routes are the mini and fast tiers, and they stop at 720p. seedance-2.0-mini-v2v at $0.0210/sec is the floor of this family; it will not deliver 1080p, so if your deliverable is 1080p this price does not exist for you.
• 2.5 costs roughly 1.5–2× the 2.0 baseline at the same tier. At 1080p: $0.5690/sec for seedance-2.5-t2v versus $0.3700/sec for seedance-2.0-t2v. You are buying longer clips (up to 30 seconds versus 4–15), mov output, and native audio generation — not more pixels.

Video-to-video is systematically cheaper than the text/image routes on the same generation: seedance-2.5-v2v starts at $0.0616/sec against $0.1030/sec for 2.5 text-to-video, and seedance-2.0-v2v undercuts 2.0 T2V at every shared tier. If your workload edits existing footage rather than generating from scratch, that gap is the single largest lever in this table. For integration specifics on the individual routes, the Seedance 2.0 API key guide and the Seedance series page cover keys and endpoints; this article stays on the money.

Seedream image pricing: one flat rate per image

Image generation is simpler: one price per output image, no resolution ladder. Five routes, two prices.

Model	Unit price	What it is for
bytedance/seedream-5.0-lite	$0.0350/img	Text-to-image with real-time web knowledge and batch output
bytedance/seedream-5.0-lite-edit	$0.0350/img	Multi-turn, reasoning-heavy image editing
bytedance/seedream-5.0-pro	$0.0810/img	Flagship text-to-image, one image per request, up to 2K
bytedance/seedream-5.0-pro-edit	$0.0810/img	Single-input precise editing (coordinates, boxes, arrows)
bytedance/seedream-5.0-pro-multi-reference	$0.0810/img	2–10 reference images blended into one output

Source: Modellix model pages, e.g. seedream-5.0-pro, seedream-5.0-lite, accessed September 13, 2026. img counts output images; the Modellix pricing-unit reference defines it that way. See the Seedream series page for the model family.

This is where the two platforms can actually be compared, because both bill per image — and the honest result is split. BytePlus/ModelArk prices seedream-5-0-pro by scenario — the single-image rate below is the one that lines up with a per-image meter, and the same page also lists scenario-specific rates for additional and layered images; that single-image rate is a pixel tier: $0.045/image at 1.5K or lower, $0.09/image above 2.61 million pixels; Seedream 5.0 Lite is a flat $0.035/image (ModelArk pricing, accessed September 13, 2026). Put side by side with the Modellix route:

• Seedream 5.0 Lite: $0.0350/image here versus $0.035/image there — the same number, no advantage either way.
• Seedream 5.0 Pro: $0.0810/image here, flat, versus $0.045 at ≤1.5K and $0.09 above 1.5K there. On small images the official tier is cheaper; on large ones our single rate is cheaper. A 1,000-image run is therefore $81.00 at a flat rate, or $45.00–$90.00 on the pixel-tiered official price depending on the mix you actually generate.

That is the shape of a real comparison — one model, one tier, one direction at a time. It is also the reason the next section matters more than any headline rate.

The per-model pricing block states the unit (Unit: $/sec) and the resolution tiers that set the rate. Screenshot captured September 13, 2026 from www.modellix.ai/models/bytedance/seedance-2.5-t2v.

The parameter that actually changes the bill

Here is the finding that no comparison page states plainly: across all 17 on-sale ByteDance models there is exactly one price dimension — resolution (on the twelve video routes; the five Seedream image routes carry no dimension at all). Not duration. Not whether audio is generated. Not how many reference images you pass, up to each model’s documented maximum. The per-model pricing block literally lists one strategy key (resolution) and its values; that is what produces the ranges in the card footers.

Three billing units, one account balance: video is charged per second of output, images per generated image, and the platform’s LLM routes per token. Illustrative Modellix artwork, not a screenshot of any ByteDance product.

What that means for cost control:

• Pick the resolution tier first, then the model. A 5-second 480p clip and a 5-second 1080p clip on the same model can differ by more than 5× — the model choice is the smaller decision.
• Longer clips do not have a higher per-second rate. Seedance 2.5 runs to 30 seconds where 2.0 stops at 15; the published rate is per second either way. A 30-second 2.5 clip is expensive because it is 30 seconds, not because long clips are surcharged.
• Inputs you send are priced as part of the route, not as extras. Reference images and audio inputs change what the model does, but the tables show no input-side surcharge line on this platform for these routes. What you pay is output seconds or output images.
• Uploaded inputs are retained for about 7 days. The File API is free and API-key-authenticated (image, video and audio; default 16 MB cap) with roughly seven-day retention — long enough for a retry loop, not long enough to be your asset store. Plan re-uploads accordingly, because losing an input reference mid-batch means paying to regenerate from a new upload.
• The model path no longer carries /async. Model invocation paths dropped the suffix on June 11, 2026; the old /async endpoints still work, so an integration written before that date will not break — but new code should call the current path. That is a migration detail, not a billing change; it is listed here because it is the one API-shape fact that shows up in every endpoint example you will copy.

Why BytePlus prices the same models differently

“Is byteplus ByteDance?” — yes. BytePlus is ByteDance’s international cloud brand, and ModelArk is its model platform, so the official route and the models discussed here come from the same company. What differs is the meter.

On the ModelArk pricing page (accessed September 13, 2026), video is billed per million tokens, with the page warning that “video pricing is determined by both the token rate and token consumption” — a higher resolution costs more because it consumes more tokens, and a video input costs more than no video input. The same page publishes per-second approximations inside its limited-time discount notices: Seedance 2.5 at 1080p at 28% off, “prices starting at approximately USD 0.41 per second” (14:00 UTC+8 August 14 through 14:00 UTC+8 September 17, 2026); Seedance 2.0 mini at 720p at 60% off, “approximately USD 0.03 per second” (August 7 through October 7, 2026); Seedance 2.0 fast at 720p at 25% off, “approximately USD 0.09 per second” (same window).

Two conclusions follow, and only two:

1. A token price and a per-second price are not comparable without the token-consumption math. Any page that slots $5.6 / M tokens next to $0.09/second as if they were the same kind of number is comparing a rate to a product of rates. Third-party routing pages that publish a single per-second figure — OpenRouter’s Seedance 2.0 page shows “from $0.06726/second”, Atlas Cloud’s ByteDance page advertises “flat $0.09/s” — are stating their own route’s number, which is legitimate, but it is one model and one configuration, not a catalog.
2. Official discounts are time-boxed and must be read as such. The 25–60% figures above are promotional windows with published end dates, not list prices. On the Modellix route none of the 17 ByteDance models carried a model-level discount on September 13, 2026 — so if a page claims a permanent “discounted ByteDance rate”, ask which window it is quoting.

None of this says which platform is cheaper overall, because the units do not line up well enough to support that claim. The comparable slice is the per-image one in the section above, and it splits both ways.

What a real job costs

Four calculations, each reproducible from the tables above. No rounding, no “typical workload” hand-waving.

Job	Calculation	Cost
10-second 1080p clip, seedance-2.5-t2v	10 × $0.5690	$5.69
Same 10 seconds at 480p on the same model	10 × $0.1030	$1.03
5-second 480p edit, seedance-2.0-mini-v2v	5 × $0.0210	$0.105
8-second 4K clip, seedance-2.0-t2v	8 × $0.7800	$6.24
1,000 images, seedream-5.0-lite	1,000 × $0.0350	$35.00
1,000 images, seedream-5.0-pro	1,000 × $0.0810	$81.00

The first two rows are the same model, the same prompt, the same duration — a 5.5× difference produced by nothing but a resolution switch. The last two rows are the same task on two tiers of the same family, a 2.3× difference. If you are estimating a video pipeline, the resolution policy is worth more engineering attention than the vendor selection.

Reconciling the bill: logs, model discovery, and CLI checks

A price list you cannot reconcile with your invoice is a marketing page. Three platform features exist specifically so each charge can be traced back to a call.

Per-request logs with cost. GET /api/v1/logs returns paginated request logs for the account that owns the API key, each entry carrying task_id, status, model.provider, model.model_name, created_at and a cost value (Get Logs, accessed September 13, 2026). The window is capped at 30 days per request and the endpoint takes an optional end-user filter, so per-customer cost attribution is a query, not an export project. Since June 2026 the logs also record the input parameters of each call — which is what makes the resolution finding above auditable: you can see the tier you were charged for, not just the total.

Token-level detail on the LLM side. The LLM log endpoint returns prompt_tokens, completion_tokens, cached_tokens and cost per request, plus latency and tokens-per-second fields (LLM logs). Media and text routes bill from different meters, and the logs keep them separate rather than blending one balance into an unexplained number.

Discovery and schema without guesswork. GET /api/v1/models lists active models with their slugs, types, documentation URLs and a description field (List models) — the programmatic equivalent of the provider page screenshotted above. The Modellix CLI adds model batch for JSONL submissions (relevant when a 500-clip run is one command, not 500 requests), and model get-schema fetches a model’s request-body contract without an API key via the public Get Schema endpoint. Cost-wise, a schema call is free; the reason to use it before a batch is that a rejected request costs nothing but a wasted run costs your time.

Two billing rules worth writing into your own runbook: on this platform sec is defined as the duration of your output video and img as the number of output images, so the meter follows what you actually receive rather than request time (pricing units); and all three product lines — media, LLM, web tools — draw from one account balance, which returns 402 when it is short rather than silently degrading.

ByteDance Model API Reference

See the request schema and pricing block for each Seedance and Seedream model before you batch.

View Docs

Top-up discounts, trial credit, and how to get a key

Three things developers ask about alongside pricing, answered from the platform’s own changelog (product updates, accessed September 13, 2026):

• First top-up: 10% off every tier. Each top-up option carries a 10% discount on your first purchase of that tier, applied at checkout. This is an account-level discount on money you add — it is not a model discount and it does not change the per-second or per-image rates in the tables above. Keep the two in separate rows of your budget.
• Signup credit is gone. New registrations no longer receive the $1 complimentary credit (ended August 19, 2026). Trial quota is requested by email from the support address on the site. Any guide promising a free starting balance for ByteDance models is describing a policy that no longer exists — and “is there a free ByteDance API tier” should be answered no until you have been granted trial quota.
• One key, one balance, both directions of usage. The key you create is the credential for these media routes and for the platform’s LLM routes; webhooks (X-Webhook-URL on prediction creation) let long video jobs push results instead of polling, and the console can email you a low-balance alert at a threshold you set — which is cheaper than discovering a 402 mid-batch.

Run ByteDance Models on One Key

Sign in to call Seedance and Seedream routes and see per-request cost in your logs.

Login

Direct BytePlus vs Modellix: when each route wins

This is a route decision, not a brand one. Both routes serve the same models; they differ in meter, activation and what you have to maintain.

Question	BytePlus / ModelArk (direct)	Modellix
Video billing unit	Per million tokens (resolution and video-input dependent)	Per second of output, per resolution tier
Image billing unit	Per image, with a pixel tier on Seedream 5.0 Pro	Per image, flat rate
Model catalog under one key	ByteDance models (plus its own LLM, 3D and embedding lines)	12 providers’ media models plus an LLM gateway, one key
Free start	Promotional windows on selected models	No signup credit; trial quota by request
Cost visibility	Platform billing console	Per-request cost + recorded input parameters, GET /api/v1/logs
What you maintain	Cloud account, region, activation and quota rules	One key and one balance

Choose direct when: you want ByteDance’s promotional windows, you need a ByteDance model we do not carry (Seedance 1.x, Seedream 4.x), you want the official platform’s support and SLA in your own cloud account, or your workload is token-shaped on ModelArk’s terms.

Choose an aggregator when: you are already calling models from several vendors and want one integration and one bill; you want per-request cost lines you can reconcile; or you are prototyping across Seedance tiers and want to switch family by changing a model slug instead of opening a second cloud account.

The thing this article does not claim is that the aggregator route is cheaper. On the one slice where both platforms bill in the same unit — Seedream image generation — the official tier is cheaper for small images and our flat rate is cheaper for large ones. Everything else is a unit mismatch that a responsible comparison refuses to paper over. If cost is your only axis, price your actual job with the table in the previous section on both routes; if integration and reconciliation are axes too, weigh them as the separate costs they are. For the same exercise run against a different provider — Qwen text, Qwen Image and Wan video — see our Alibaba API pricing breakdown; the units change, the method does not.

A pricing checklist before you commit

• Name the model and the resolution. Twelve of these seventeen routes change price with resolution; a quote without a tier is not a quote.
• Check the unit before you compare anything. sec, img, and M tokens are not interchangeable; only same-unit figures can be placed side by side.
• Ask whether a discount is promotional or permanent. Official windows have end dates; this platform’s ByteDance rates carried no model-level discount on September 13, 2026.
• Keep account-level discounts separate. The 10% first top-up is money off your balance, not off the rate.
• Budget retries, not just passes. The meter follows your output — sec of delivered video, img of delivered images — so a quality gate that rejects one in five generations still costs you the reruns you generate, and File API inputs expire after roughly seven days.
• Read one invoice line before you scale. Pull a request log, match the cost to the model and tier you sent, and confirm the meter behaves as documented. Ten minutes now beats discovering a 4K mistake on the next thousand clips. The public pricing table lists live per-model rates across every provider on the platform if you want to widen the comparison beyond this family.

Frequently Asked Questions

How much does ByteDance API pricing cost per video?
On the Modellix route, Seedance video runs from $0.0210/second (seedance-2.0-mini-v2v at 480p) to $0.7800/second (seedance-2.0-t2v or -i2v at 4K), all rates as of September 13, 2026. The rate is set by resolution: seedance-2.5-t2v is $0.1030/sec at 480p, $0.2310/sec at 720p and $0.5690/sec at 1080p.

What is the cheapest Seedance 2.0 API version?
Among the twelve Seedance routes on sale today, the lowest published rate is bytedance/seedance-2.0-mini-v2v at $0.0210/second at 480p ($0.0450 at 720p). It is the cheapest by design — a lightweight video-to-video tier capped at 720p. If your job is text-to-video rather than video editing, the floor is $0.0400/second (seedance-2.0-mini-t2v, 480p).

How much does Seedance 2.5 cost?
Seedance 2.5 has three routes on sale: -t2v and -i2v at $0.1030/sec (480p), $0.2310/sec (720p), $0.5690/sec (1080p), and -v2v at $0.0616/sec, $0.1380/sec and $0.3400/sec for the same three tiers. A 10-second 1080p clip on seedance-2.5-t2v is therefore $5.69. Seedance 2.5 supports clips up to 30 seconds and mp4 or mov output.

Is BytePlus ByteDance?
Yes. BytePlus is ByteDance’s international cloud brand and ModelArk is its model platform, so the official route serves the same Seedance and Seedream models described here. The difference is the meter: ModelArk bills video per million tokens and images by pixel tier, while Modellix lists a per-second rate for video and a flat per-image rate.

How much does Seedream 4.0 cost?
Not from Modellix — Seedream 4.x is not on sale here as of September 13, 2026; the Seedream line starts at 5.0 ($0.0350/image for Lite, $0.0810/image for Pro). ByteDance’s own platform lists seedream-4-0-250828 at $0.03/image on the ModelArk pricing page (accessed September 13, 2026), so that is the number to use if you need 4.0 specifically.

Is there a free ByteDance API tier?
No. The $1 signup credit ended on August 19, 2026, and there is no standing free allowance for these models. Trial quota is requested by email from the support address listed on the platform; until you have been granted it, budget from the paid rates above.

Does resolution or duration change the ByteDance API price more?
Resolution, and by a wide margin. Duration simply multiplies the per-second rate — a 30-second clip costs three times a 10-second clip at the same tier. Resolution changes the rate itself: on seedance-2.0-t2v, 4K is $0.7800/sec against $0.0700/sec at 480p, more than an 11× difference on the same model and prompt.

Why do different websites quote different ByteDance API prices?
Because they quote different units and different routes. A per-million-token figure (ModelArk’s official video billing), a per-second figure for one model under one configuration (routing platforms), and a per-image figure are three different meters — and promotional windows add a fourth variable, since official discounts of 25–60% carry published end dates. Compare only same-unit figures, and always check the date attached to a number.

---

ByteDance model prices and platform capabilities reflect the Modellix model pages and ByteDance/BytePlus official pricing pages accessed September 13, 2026; both change without notice, and per-model promotional windows on the official platform expire on published dates. Validate against the live model pages and each provider’s current pricing before committing to a workload. Modellix is an API aggregator for image and video models and has a commercial interest in the routes it resells — it does not train or own the models described here. Access Seedance, Seedream and other providers’ image and video models through a single API key at modellix.ai.