BytePlus ModelArk gives developers and enterprises pay-as-you-go API access to Seed language models, Seedance video and Seedream image generation, plus third-party LLMs, agent tooling and a subscription for AI coding tools.
Used this tool? Rate it
Used this tool? Rate it
BytePlus is an enterprise AI cloud that sells model APIs, agent services and compute to businesses and developers. Its homepage pitch is to build, deploy and scale AI with models, tools and infrastructure that the company says are proven on global platforms.
The company says it was born from ByteDance's technology. Its Terms of Service name BytePlus Pte. Ltd., a company incorporated in Singapore, as the provider of the services.
The AI line centers on BytePlus ModelArk, which BytePlus calls a one-stop large language model service platform. Through one ModelArk account you can call ByteDance's Seed language models, Seedance video models and Seedream image models alongside third-party models such as DeepSeek and GLM, paying per token, per image or per hour of dedicated capacity.
Beyond chat and reasoning, ModelArk's document understanding accepts PDFs and replies based on their content. Web search lets models obtain real-time knowledge from the internet, and function calling connects them to your own tools.
Dreamina Seedance 2.5, the mainline model behind the Seedance API, supports native 30-second video generation and up to 50 total reference assets per generation across images, videos and audio.
Only successfully generated videos are billed, so a request blocked by content moderation costs nothing.
BytePlus positions its infrastructure as an end-to-end cloud for AI, from training to serving, with built-in security and control.
For Seedance 2.5 and 2.0, the documentation offers two routes for portraits. A preset digital character library supplies portrait assets for scenes that need a realistic face without naming a specific person. For a named person, their image, video and audio can be uploaded to ModelArk after real-person verification and that person's authorization, and each stored asset receives its own asset ID for generation calls.
BytePlus suits buyers who want ByteDance's models through an official enterprise channel, even when the rest of the stack lives elsewhere.
It is a weaker fit for buyers outside the supported country list, for workloads that need fine-tuned model weights exported, and for anyone expecting refundable monthly subscriptions.
ModelArk is used through its API and web console, and currently runs in two regions:
| Region | Region ID | Models |
|---|---|---|
| Johor, Asia Pacific | ap-southeast-1 | Every model in the model list |
| Dublin, Europe | eu-west-1 | A subset, listed inconsistently (see below) |
The EU model list is inconsistent between pages. The region page says the EU region currently supports seed-2-0-lite, while the model list says the seed-2-0 and seedream-5-0-lite models are also supported in eu-west-1.
Country coverage is narrower than the region list suggests. Apart from Restricted Models, BytePlus Model Service is available only in a published list of countries and regions, and availability is determined at the point of purchase. The United States does not appear on the list captured on September 28, 2026, which runs from the United Kingdom straight to Uruguay. India is also absent, with Iceland followed directly by Indonesia. Support is offered in Simplified Chinese, Traditional Chinese and English, and documentation articles are in English.
For AI coding, the Coding Plan works with Claude Code, OpenCode, OpenClaw, TraeCode, Cline, Cursor, Codex, Roo Code and Kilo Code, and quota is shared across those tools. It exposes both an OpenAI-compatible and an Anthropic-compatible base URL.
Features and unit prices vary with model services, so there is no single ModelArk price sheet. The homepage advertises free image and video tokens to experience models and headline prices as low as 0.03 USD per image and 0.05 USD per video. All figures below are list prices in USD as captured on September 28, 2026, and the availability terms note that model versions can differ by country or region.
| Model | Input | Output | Condition |
|---|---|---|---|
| seed-2-0-pro | 0.50 | 3.00 | Prompts up to 128K tokens |
| seed-2-0-mini | 0.10 | 0.40 | Prompts up to 128K tokens |
| glm-5-2 | 1.4 | 4.4 | Single tier |
| deepseek-v4-1-flash | 0.15 | 0.60 | Off-peak hours |
| deepseek-v4-1-flash | 0.30 | 1.20 | Peak hours |
Longer prompts move Seed models into a higher tier. DeepSeek V4.1 Flash peak hours are Monday to Friday, 09:00–12:00 and 14:00–18:00 (UTC+8), and the console displays the peak price. Flex online inference charges roughly 50% of standard input and output prices, and batch inference also runs at lower rates, for example 0.125 USD input and 1.00 USD output per million tokens for seed-2-0-lite-260228.
Seedance 2.5 costs 10.70 USD per million tokens for 480p and 720p output without input video, and 11.7 USD per million tokens at 1080p without input video; offline inference is not supported yet. Estimated token use for a clip is (input video duration + output video duration) × width × height × frame rate ÷ 1024, and the billed figure is the usage the API returns. Until 14:00 (UTC+8) on October 7, 2026, Seedance 2.0 mini is 60% off list and 2.0 fast 25% off list for enterprise users.
Seedream 5.0 pro charges 0.045 USD per output image up to about 2.61 million pixels and 0.09 USD above that. Its first input image is free and each further input image costs 0.003 USD. Seedream 5.0 flash costs 0.018 USD per image and Seedream 4.0 costs 0.03 USD.
| Coding Plan | 1 month | 3 months |
|---|---|---|
| Lite | 10 USD | 30 USD |
| Pro | 50 USD | 150 USD |
Lite allows up to about 1,900 requests every 5 hours, 12,000 per week and 24,000 per subscription month; Pro raises that to about 9,500, 60,000 and 120,000. A single prompt in a coding tool often triggers several model calls, so requests run out faster than prompts. Once purchased, a Coding Plan subscription cannot be refunded, even if the quota goes unused. Automatic renewal has to be cancelled at least 7 days before the next billing date to avoid the renewal charge.
Managed Agents add 0.08 USD per hour of running sandbox time and 0.006 USD per web search on top of model tokens, while web fetch is free for a limited time. Dedicated model units cost 4.20 USD per hour or 2800.00 USD per month for the Flavor-A machine type, and the feature is a beta open only to invited users.
It is built on ByteDance technology and sells ByteDance's models, but the contracting entity is BytePlus Pte. Ltd. in Singapore.
Free image and video tokens are offered for trying models, and activating a model costs nothing until it is used; otherwise usage is pay-as-you-go.
No. The quota works only inside supported AI coding tools; apps need the pay-as-you-go model APIs.
Not unless you opt in, although content that triggers the safety filter is kept for 180 days.
