ByteDance’s Volcano Engine has raised its full-year model-as-a-service (MaaS) revenue target to RMB 15 billion (USD 2.2 billion) after Seedance 2.0 became a major contributor to growth.
According to 36Kr, the latest revision was made in April and followed a series of upward adjustments made almost every month. At the end of 2025, the target was reportedly RMB 10 billion (USD 1.5 billion).
Volcano Engine’s full-year MaaS revenue in 2025 was around RMB 1.5 billion (USD 221 million), meaning its latest 2026 target is about ten times last year’s actual revenue. Much of that growth can be traced to the rapid adoption of Seedance 2.0, the video model now generating more than RMB 1 billion (USD 147.3 million) in monthly revenue.
A person close to Volcano Engine said its average daily token consumption has continued to grow by nearly 40% month-on-month.
That points to a clearer commercial pattern: strong artificial intelligence models can now drive higher token consumption and convert more users into paying customers.
Seedance 2.0 becomes the growth driver
Competition for token usage has split into two main fronts: coding and video generation. In conventional text chat, companies have pushed model prices close to the floor. Coding and video generation still retain some pricing power, and ByteDance is benefiting from the latter through Seedance 2.0.
After ByteDance released Seedance 2.0 on February 10, the model quickly gained traction globally. It led the Artificial Analysis Video Arena rankings with an Elo score of 1,269 and surpassed leading alternatives, including Google’s Veo 3 and OpenAI’s Sora 2, across several metrics.
Its popularity also created shortages. During the Lunar New Year period, users at one point faced long queues, with videos just over ten seconds long taking as long as ten hours to generate. ByteDance later introduced measures including whitelists, giving customers with annual contracts stable access to the full-capability version of Seedance 2.0. That marked the beginning of the model’s commercialization.
Several sources from major tech companies told 36Kr that Seedance 2.0 alone can generate more than RMB 1 billion in monthly revenue for Volcano Engine, and that the figure is still rising. They expect Volcano Engine to continue raising its MaaS revenue target.
Seedance 2.0 has validated a simple point: when a model’s capabilities are clearly differentiated, paying customers tend to follow. Seedance’s penetration rate in the short drama industry has reportedly reached about 95%. “If you work in short dramas, it is hard not to be a Seedance user,” one video agent practitioner told 36Kr.
Seedance 2.0 currently ranks second globally by market share, behind only Google’s Veo. This year, its goal is to expand further overseas and become the world’s largest video model by market share. If that effort succeeds, Volcano Engine’s MaaS revenue is likely to keep rising.
ByteDance lags in AI coding
Not every model sold by Volcano Engine is performing as strongly.
Coding has not brought Volcano Engine much incremental revenue. The reason is straightforward: ByteDance Seed’s coding capabilities are not yet strong enough.
In coding, more than one person familiar with the matter told 36Kr that Zhipu AI, also known as Z.ai, may be the biggest winner.
In 2024, Zhipu bet on a training path similar to Anthropic’s, focusing on model intelligence and coding capabilities. Two years of accumulated work gave Zhipu an opportunity to take a leading position.
Zhipu CEO Zhang Peng disclosed on an earnings call that GLM’s API call prices increased by about 83% in aggregate in the first quarter of 2026. Despite several price adjustments, API call volume still grew 400%. The cached-hit token price for Zhipu’s GLM-5.1 is now close to the level of Claude Sonnet 4.6, marking the first time a Chinese-built large model has matched leading overseas providers on price in a core use case. Behind that pricing is GLM-5.1’s leading domestic position in coding capability.
Compared with video models, coding is already a more crowded revenue track.
At the end of 2025, API service provider OpenRouter proposed that early users who adopt an AI model in its first month after launch often show higher retention than those who arrive later. When users find an effective model, they tend to build loyalty around it and embed it deeply into their workflows.
As a result, Moonshot AI, MiniMax, cloud services from major companies such as ByteDance, Alibaba, and Tencent, and even telecom operators have been rolling out coding plans and increasing sales efforts. Their aim is to attract users with relatively low prices, bring them into their ecosystems, generate large volumes of token consumption, and then find ways to improve margins.
36Kr has learned that Alibaba Cloud has set up several new MaaS sales teams since last year. These teams can work with its existing direct sales teams on deals. Their main task is to identify use cases, push enterprise customers to adopt agent capabilities, and increase token consumption.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Deng Yongyi for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.79 = USD 1 based on estimates as of June 4, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.