On June 23, Volcano Engine hosted the 2026 Summer FORCE Conference in Beijing, centrally releasing the Doubao Large Model 2.1, along with several new models for video, image, and audio, and upgrading its cloud service system for Agents.
According to Volcano Engine, as of June this year, the average daily token call volume for the Doubao large model has exceeded 180 trillion, a more than tenfold increase over the past year. IDC data shows that in China's public cloud MaaS service market, Volcano Engine ranks first with a 49.5% market share.
Doubao 2.1 Pro Surpasses the Production-Grade "Tipping Point"
Tan Dai, President of Volcano Engine, stated that a model must cross a "tipping point" in capability to truly meet the needs of enterprises and individuals in production scenarios. Globally, the first video generation model to cross this tipping point was Seedance 2.0, while in the Coding and Agent fields, it was Claude Opus 4.6.
As the flagship model released at this conference, Doubao 2.1 Pro has achieved a leap in capability across three core directions: Coding, Agents, and VLM (Vision Language Models), with multiple evaluation results surpassing Claude Opus 4.6, officially crossing the production-grade tipping point.
Coding Capability Evaluation
Agent Capability Evaluation
VLM Capability Evaluation for GUI, Image Understanding, etc.
Evaluation results show that Doubao 2.1 Pro has entered the top tier in code evaluations such as Terminal Bench 2.1, SWE-Pro, and SciCode. In terms of Agents and multimodal capabilities, the model also ranks among the global leaders in evaluations like OSWorld, MobileWorld, and MMMU-Pro.
Tan Dai demonstrated the model's Coding and Agent capabilities with practical cases: In a chip design RTL test, Doubao 2.1 Pro ran continuously for nearly 18 hours, underwent 9 rounds of iteration, and successfully completed the entire engineering process including simulation, testing, and synthesis checks, demonstrating the model's production-grade Coding delivery capability in a real engineering scenario. Using the Doubao 2.1 Pro model to build a 3D virtual city scene, it enabled over 500 intelligent Agents to collaborate synchronously, completing over a thousand tool calls and generating more than a hundred buildings.
It is reported that Doubao 2.1 Pro is priced at 6 yuan per million input tokens and 30 yuan per million output tokens, with a cache hit price of only 1.2 yuan. Volcano Engine states its comprehensive usage cost is nearly 80% lower than Claude Opus 4.6. For high-frequency call scenarios, Doubao 2.1 Turbo is priced at half the cost of 2.1 Pro, offering even better cost-effectiveness. Meanwhile, Volcano Engine launched the Doubao-Seed-Evolving version, which will rapidly iterate at a frequency of 2 to 4 times per month for Coding and Agent scenarios.
Currently, the Doubao 2.1 model is available via Volcano Engine's API service and is gradually being integrated into products like Doubao, TRAE, and Coze.
Seedance 2.5 Unveiled, Expanding Application in Physical Industries
The latest Doubao video generation model, Seedance 2.5, made its debut at the FORCE conference and is expected to officially launch in July.
According to the introduction, Seedance 2.5 achieves three capability enhancements: direct output of 30-second single-segment native videos, support for joint generation with up to 50 full-modal materials, and localized editing that maintains picture consistency.
Tan Dai believes that video generation is one of the paths to world models and has immense potential for application in physical industries. Currently, Seedance has been deployed in fields such as embodied intelligence, industrial manufacturing, and intelligent driving, providing new tooling capabilities for business needs like data synthesis, scenario simulation, and process demonstration.
Furthermore, Volcano Engine showcased the image creation model Seedream 5.0 Pro and the audio generation model Seed-Audio 1.0, further consolidating its multimodal leading advantage.
Seedream 5.0 Pro features capabilities such as interactive precise editing, multi-layer separation, high-density information expression, and native multilingual text generation, enabling the direct generation of editable layered design drafts and complex infographics. Seed-Audio 1.0 features zero-shot multimodal reference capability, supporting the simultaneous generation of multi-character dialogues, background music, and Foley sound effects.
For the Agent era, Volcano Engine once again upgraded its AI cloud-native architecture, releasing the Ark CLI command-line tool, which supports developers in connecting Agents to Volcano Ark with a single command. It also upgraded AgentKit and HiAgent 3.0, and launched the ArkClaw Enterprise Edition Agent workbench and AI Trust product system, completing the infrastructure and security capabilities for enterprise Agent applications.
The conference disclosed that over 1.1 million enterprises and individuals currently use the Volcano Ark large model service. The number of enterprises with an annual token call volume exceeding 1 trillion has reached 200, doubling in half a year, covering multiple industries including internet, manufacturing, finance, and automotive.
END