SenseTime today (the 11th) announced that the new multimodal agent model SenseNova 6.8 Flash Lite Preview is now officially available on the SenseNova Token Plan.
The model focuses on lightweight agility, with capabilities including autonomous planning, continuous execution, multi-Agent collaboration, and dynamic error correction. It can deliver complete end-to-end results and is evolving toward Delegated Intelligence: users only need to provide a goal, AI autonomously completes the entire process, and people focus on reviewing the results.
IT Home learned that the official 6.8 Flash Lite model, along with the higher-performance 6.8 Flash model, will also be released in the near future.
According to the introduction, SenseNova 6.8 Flash Lite Preview delivers breakthrough enhancements in three key capabilities centered on the requirements of completing long-horizon, complex tasks:
Long-form and long-horizon stability: It can maintain goals, constraints, and key facts during long-horizon tasks spanning hundreds of steps, multiple phases, and several hours, while autonomously rolling back and replanning when failures occur.
Dynamic Sub-agent collaboration: The main Agent can dynamically schedule and organize more than a dozen specialized Agents to perform information retrieval, data analysis, mathematical calculations, visual understanding, and fact verification in parallel.
Native multimodal integration: It enables integrated planning, reasoning, tool use, and result verification for text, images, charts, documents, video, and application interfaces within the same task trajectory.
The model can create dynamic PPTs, handling content writing, chart creation, and layout design. It supports native HTML dynamic presentations, claiming to overcome the limitations of traditional static layouts; it can natively understand multimodal information on web pages, application interfaces, and local files, directly operate browsers and applications on your behalf, and autonomously complete office tasks, script writing, and automated execution.
The model is now officially available for hands-on use at the following links:
SenseNova Token Plan: https://www.sensenova.cn/token-plan
GitHub repository: https://github.com/OpenSenseNova/SenseNova6.8