ByteDance used its Volcano Engine FORCE conference to introduce a new wave of AI products, led by Doubao 2.1 Pro and a preview of Seedance 2.5, its next major video generation model. The company said Doubao 2.1 Pro has reached what it describes as a production-ready capability level, while Seedance 2.5 is scheduled to launch in July.
According to ByteDance, Doubao 2.1 Pro brings major improvements in three areas: software coding, agent performance, and visual language understanding. The company said the model posted strong results on several benchmarks, including Terminal Bench 2.1, SWE-Pro and SciCode for code generation. It also said the model ranked highly on OSWorld, MobileWorld and MMMU-Pro, which test agent behavior and multimodal reasoning.
ByteDance further claimed that Doubao 2.1 Pro outperformed Claude Opus 4.6 in multiple evaluations. The company paired those performance claims with aggressive pricing. It said the model costs 6 yuan per million input tokens and 30 yuan per million output tokens, while cache-hit pricing can fall to 1.2 yuan. Volcano Engine said the total cost of ownership is about 80 percent lower than Claude Opus 4.6.
A lower-cost Turbo version is also available for heavier workloads, priced at half the rate of the standard Pro model.
ByteDance said usage of the Doubao family has scaled quickly, with daily token calls exceeding 180 trillion in June. The company said that is more than 10 times higher than a year earlier. It also cited IDC data showing Volcano Engine holding 49.5 percent of China’s public cloud MaaS, or Model as a Service, market.
The company’s video roadmap centered on Seedance 2.5, which ByteDance positioned as a substantial upgrade over earlier versions. The model is expected to generate native 30-second videos in a single pass and can work with as many as 50 multimodal assets in joint generation.
ByteDance said the new model adds local editing tools while preserving consistency across frames, a key challenge in AI video systems. Volcano Engine president Tan Dait said video generation could become an important step toward world models and pointed to potential industrial uses. The company said Seedance is already being used in live-stream e-commerce, industrial manufacturing and autonomous driving, where it supports data synthesis and simulation.
Alongside the flagship models, ByteDance introduced Seedream 5.0 Pro for image generation. The company said the system supports interactive editing and multi-layer separation. It also previewed Seed-Audio 1.0, which can generate audio with multiple characters, background music and sound effects in zero-shot mode.
On the infrastructure side, Volcano Engine launched Ark CLI, a tool for one-command agent deployment, and said it upgraded AgentKit and HiAgent 3.0. The company also introduced ArkClaw, an enterprise-grade agent workbench designed for business users.
ByteDance said more than 1.1 million enterprises and individual developers now use the Volcano Ark model service. It added that 200 companies have each surpassed 1 trillion annual token calls, underscoring the scale of adoption the company is seeing across its AI platform.
The announcements suggest ByteDance is pushing to compete not only on model quality, but also on pricing, deployment tools and sector-specific applications as it expands its AI ecosystem.