BytePlus vs Volcengine: ModelArk vs Ark, models, prices and which one to use (2026)
BytePlus is ByteDance's international cloud; Volcengine serves China. How ModelArk and Ark differ on Seed (Doubao) models, USD vs CNY prices and regions.
MofCloud
Nov 19, 2026 · 7 min read
BytePlus and Volcengine are both ByteDance’s enterprise clouds: Volcengine (火山引擎) serves mainland China, and BytePlus serves everyone else. Their model platforms mirror each other. Volcengine Ark (火山方舟) sells ByteDance’s Doubao Seed models in CNY from China, and BytePlus ModelArk sells Seed models in USD from Malaysia and Ireland. If your company and users are outside China, BytePlus is the one you can sign up for. The two aren’t price-identical, though: as of November 19, 2026, some Seed models cost about three times as much on BytePlus, while some long-prompt tiers cost less.
This post compares the two platforms on models, prices, billing rules and regions, using each one’s official price list.
What’s the difference between BytePlus and Volcengine?
| Volcengine Ark (火山方舟) | BytePlus ModelArk | |
|---|---|---|
| Market | Mainland China | International |
| Console and docs | Chinese | English (plus other languages) |
| Currency | CNY | USD |
| Regions | Mainland China | Johor, Malaysia (ap-southeast-1); Dublin (eu-west-1) for some models |
| API base URL | ark.cn-beijing.volces.com |
ark.ap-southeast.bytepluses.com, ark.eu-west.bytepluses.com |
| Model naming | doubao-seed-* |
seed-*, with the newest as dola-seed-* |
| ByteDance models | Doubao Seed 2.1 Pro, Lite and Turbo; Seed 2.0 Pro, Lite, Mini and Code | Dola Seed 2.1 Turbo; Seed 2.0 Pro, Lite, Mini and Code |
| Third-party models | DeepSeek, GLM, Kimi | DeepSeek, GLM |
| Inference modes | Standard, low-latency, low-priority, Batch, TPM guarantee packages | Standard, Fast, Flex, Batch |
Sources: Volcengine Ark model pricing and BytePlus ModelArk model list and region availability, as of November 19, 2026.
In one line: same company, same model family, two storefronts, and the storefront you can use depends on where your company is.
Which Seed (Doubao) models are on each platform?
Volcengine gets ByteDance’s full lineup. BytePlus carries a subset, and the newest flagship isn’t on it yet:
- Doubao Seed 2.1 Pro is listed on Volcengine (¥6 input / ¥30 output per million tokens) but not on BytePlus’s model list. See its price page.
- Seed 2.1 Turbo is on both:
doubao-seed-2.1-turboon Volcengine anddola-seed-2-1-turboon BytePlus, which features it as its general-purpose agent model with a 256K context. - Seed 2.0 Pro, Lite, Mini and Code are on both.
- Kimi K3.1 Pro is hosted on Volcengine but not on BytePlus.
- DeepSeek V4 Pro and V4.1 Flash, and GLM-5.2 and GLM-5.3-Flash, are on both.
BytePlus’s region availability page says every model is available in Johor, while the Dublin region currently lists seed-2-0-lite and supports online inference but not batch inference.
BytePlus vs Volcengine pricing: which is cheaper?
It depends on the model and the prompt length. Here are standard online-inference prices per million tokens, as of November 19, 2026. Volcengine prices are converted at 6.6672 CNY per USD (ECB rate, November 18, 2026), so treat small differences as noise.
| Model | Prompt length | Volcengine (CNY → USD) | BytePlus (USD) | BytePlus vs Volcengine |
|---|---|---|---|---|
| Seed 2.1 Turbo | up to 256K | ¥3 / ¥15 → $0.45 / $2.25 | $0.50 / $2.50 | About 11% higher |
| Seed 2.0 Pro | up to 32K | ¥3.2 / ¥16 → $0.48 / $2.40 | $0.50 / $3.00 | Similar input, 25% higher output |
| Seed 2.0 Pro | 128K–256K | ¥9.6 / ¥48 → $1.44 / $7.20 | $1.00 / $6.00 | About 30% lower input, 17% lower output |
| Seed 2.0 Lite | up to 32K | ¥0.6 / ¥3.6 → $0.09 / $0.54 | $0.25 / $2.00 | About 3–4x higher |
| Seed 2.0 Mini | up to 32K | ¥0.2 / ¥2 → $0.03 / $0.30 | $0.10 / $0.40 | About 3x higher input |
| DeepSeek V4 Pro | — | ¥9 / ¥27 → $1.35 / $4.05 | $1.32 / $3.96 | About the same |
| DeepSeek V4.1 Flash (peak) | — | ¥2 / ¥8 → $0.30 / $1.20 | $0.30 / $1.20 | Same |
| GLM-5.2 | — | ¥8 / ¥28 → $1.20 / $4.20 | $1.40 / $4.40 | About 5–17% higher |
Prices are input / output. Sources: Volcengine Ark model pricing (updated November 19, 2026) and BytePlus ModelArk model pricing. For today’s numbers, see our ByteDance (Doubao) provider page and the LLM price comparison, which sync daily.
Three patterns stand out:
- The small Seed models cost much more on BytePlus. Seed 2.0 Lite and Mini are about 3x the Volcengine price for short prompts. If you picked them because they’re cheap in Chinese benchmarks, re-run the numbers in USD.
- The prompt-length tiers are cut differently. Volcengine prices Seed 2.0 in three bands (up to 32K, 32K–128K, 128K–256K). BytePlus uses two (up to 128K, 128K–256K). A 100K-token prompt to Seed 2.0 Pro pays Volcengine’s middle band (¥4.8 / ¥24, about $0.72 / $3.60) but BytePlus’s lowest ($0.50 / $3.00). For long-context work, BytePlus can be the cheaper of the two.
- Third-party models track their makers’ own prices. DeepSeek on BytePlus costs exactly what DeepSeek charges directly, and GLM-5.2 on BytePlus matches Z.ai’s $1.40 / $4.40. There’s no discount for buying through ByteDance, but no markup either.
Our take: compare BytePlus with other international options, not with Volcengine. A team outside China can’t realistically buy from Volcengine, so the useful question is whether Seed 2.0 Lite at $0.25 / $2.00 beats, say, Qwen Flash or GLM-5.3-Flash for your task. Often it won’t on price alone; pick it for multimodal input or if you already run on BytePlus.
How do billing rules differ?
Most rules are shared, since both platforms run the same Ark software:
- Cache storage fee. Both charge for cached tokens by the hour on top of a discounted cache-hit price: ¥0.017 per million tokens per hour on Volcengine and $0.0083 on BytePlus. A large, long-lived cache isn’t free.
- Tiered pricing by prompt length. On both, the tier is set by each request’s prompt length and applies to the whole request.
- Off-peak pricing on DeepSeek V4.1 Flash. Both charge half price outside peak hours, defined on both as 09:00–12:00 and 14:00–18:00 Beijing time on weekdays, which matches DeepSeek’s own schedule.
- Cheaper tiers. BytePlus sells Flex, at about 50% of Standard input and output with the same cached-input price, and Batch. Volcengine also has Batch, plus low-latency and low-priority online modes. Batch prices on both are roughly half of Standard.
Regions and data residency on BytePlus
BytePlus says ModelArk may use data centers in Malaysia, Indonesia and/or the EU/EEA, and that it won’t use customer data to train its own models without authorization (BytePlus ModelArk data processing). Two details matter for compliance:
- Requests can spill across regions. BytePlus says traffic to an endpoint goes primarily to its region, but “some requests may be routed to inference resources in other regions.” An EU endpoint doesn’t guarantee EU-only processing.
- The US isn’t on the country list. BytePlus’s model service availability list, last updated April 21, 2026, doesn’t include the United States or mainland China.
Volcengine data stays in mainland China, which is the point for Chinese customers and the problem for most others. We cover cross-border rules in Is your cloud bill safe with a FinOps tool?.
BytePlus or Volcengine: which should you use?
- Your company is in mainland China, or your users are: Volcengine. You get the full model lineup, CNY billing, lower prices on small models, and data stays in China.
- You’re outside China and want Seed models: BytePlus ModelArk. It’s the international route ByteDance offers, and the only one of the two most non-Chinese companies can sign up for.
- You’re outside China and only want DeepSeek or GLM: BytePlus is a reasonable host at the makers’ own prices, with inference in Malaysia or Ireland. Weigh it against DeepSeek’s own API, Z.ai and Alibaba Cloud. Our guide to using Chinese AI APIs from outside China compares the routes.
- You expand overseas from China and serve both markets: you’ll likely run both, with two bills in two currencies.
That last case is where costs get hard to read. A Doubao line on a Volcengine bill in CNY and a Seed line on a BytePlus invoice in USD are the same model family, but nothing on either bill tells you that. Mofis, our desktop app, reads Volcengine’s bill, breaks Ark spend down by model and key, and shows it next to Alibaba Cloud, Tencent Cloud, AWS, Azure and Google Cloud in one currency. For how the token-level side of this works, see LLM API cost tracking, and for how Chinese model prices compare overall, Chinese LLM API pricing in 2026.
FAQ
Is BytePlus the same as Volcengine?
Same parent, different clouds. Both belong to ByteDance. Volcengine serves mainland China with CNY billing; BytePlus serves international customers in USD. Their AI platforms, Volcengine Ark and BytePlus ModelArk, share a design but have separate accounts, model lists and price lists.
Is Doubao available on BytePlus?
ByteDance’s Seed models, sold as Doubao in China, are available on BytePlus ModelArk under seed-* and dola-seed-* names. As of November 19, 2026, Seed 2.1 Turbo and the Seed 2.0 family are listed; Doubao Seed 2.1 Pro is listed on Volcengine only.
Is BytePlus cheaper than Volcengine?
Usually not. Small Seed models cost about 3x more on BytePlus for short prompts. Exceptions are long prompts, where BytePlus’s wider lowest tier can make Seed 2.0 Pro cheaper, and DeepSeek models, which cost the same on both.
What is BytePlus ModelArk’s API base URL?
https://ark.ap-southeast.bytepluses.com/api/v3 for Johor (ap-southeast-1) and https://ark.eu-west.bytepluses.com/api/v3 for Dublin (eu-west-1). API keys and model activation are separate per region.
Can I use BytePlus ModelArk from the United States?
BytePlus’s published availability list for its model service, last updated April 21, 2026, doesn’t include the United States. Check the current list before you build on it.
Running Doubao on Volcengine and other models elsewhere? Mofis shows all your model and cloud spend in one view. Free for one person, with a 14-day Pro trial and no card required.