SenseTime Releases SenseNova 6.7 Flash-Lite, Cutting Token Consumption by 60% and Opening Token Plan

A lightweight native multimodal agent model for real-world workflows

SenseTime has released SenseNova 6.7 Flash-Lite, a new lightweight multimodal agent model. At the same time, the SenseNova Token Plan is available for a limited-time free trial, and the full SenseNova-Skills office skill suite has been open-sourced on GitHub.

Users can log in at https://www.sensenova.cn/ to claim 1,500 free calls every five hours, making intelligent office workflows faster, better, and more cost-efficient.

SenseNova 6.7 Flash-Lite: designed for workflows, with 60% lower token consumption

Conventional agent models usually stitch language and vision together, treating vision as an auxiliary text input. As a result, visual information is damaged during translation, while token consumption becomes expensive in long-chain tasks.

SenseNova 6.7 Flash-Lite changes this path. Built on a native multimodal architecture, it directly understands complex web layouts, document structures, financial charts, and other real-world inputs, unifying seeing, thinking, and acting in one workflow. It improves the success rate of data analysis, deep research, PPT generation, and other long-chain complex tasks.

By removing the visual-to-text intermediate layer, the model achieves stronger agent capabilities with a smaller parameter scale. It reaches multiple SOTA results among models at the same level on authoritative agent benchmarks, and reduces token consumption by up to 60% in information-search scenarios compared with text-only agents. Millisecond-level feedback also makes it suitable for high-frequency production environments.

Technical details are available at https://github.com/OpenSenseNova/SenseNova6.7.

SenseNova Token Plan: limited-time free access

SenseNova Token Plan is now available with a limited-time free offer. Developers can receive 1,500 calls every five hours in the first month, unlocking a zero-cost, high-speed AI experience.

The plan is powered by SenseTime's model stack: SenseNova-Skills for high-frequency office scenarios, SenseNova 6.7 Flash-Lite for complex data analysis and task planning, and SenseNova U1 Fast for infographic generation.

Through a more efficient model portfolio and more generous token quotas, the plan reduces cost concerns for developers and enterprise users, allowing long-running tasks to run reliably. Developers can apply at https://www.sensenova.cn/. More Lite and Pro tiers will be introduced for creators and enterprises of different scales.

The plan is ready to use without complex configuration, helping developers quickly integrate professional office AI skills and accelerate agile development.

One-click automated office workflows with SenseNova-Skills

Complete workflows require deep collaboration between an agent framework and professional skills. SenseTime has packaged the core capabilities of the SenseNova model family into SenseNova-Skills, covering infographic generation, PPT creation, data analysis, deep research, and other high-frequency office scenarios.

SenseNova-Skills natively supports agent frameworks such as OpenClaw and Hermes Agent. Users can call individual skills as needed or combine them into end-to-end complex workflows.

Recommended path: use the Agent Pack at https://github.com/OpenSenseNova/agent_pack. It integrates the Hermes Agent or OpenClaw framework with the full Skills plug-in set and can be used with the free SenseNova Token Plan. Users of other agent frameworks can directly install components from https://github.com/OpenSenseNova/SenseNova-Skills.

Case study: from data insight to professional delivery

SenseNova 6.7 Flash-Lite is not only capable of reasoning; it is also strong at producing high-quality deliverables and closing the loop from complex document parsing to structured output.

In a sales analysis task for a chain enterprise, the model processed 36 months of sales records with nearly 900,000 rows. It followed a complete workflow of data processing, data analysis, visual design, conclusion summarization, and optimization recommendations, producing an operational data analysis and decision-support report.

The model first performed a data audit instead of jumping directly into statistics. It detected abnormal unit prices and reasoned that those outliers reflected specific promotion or premium-product scenarios, so it retained them to reflect real market volatility.

When analyzing gross margin, the model found severe negative gross margin for chili products in May 2022. It then explored the supply-chain links and diagnosed the lack of coordination between procurement cost control and retail pricing. Based on the insights, it proposed dynamic pricing and category-structure adjustment recommendations to support management decisions.

Case study: embodied intelligence industry research

In a deep research task, SenseNova 6.7 Flash-Lite independently completed an eight-chapter market research report on the current state of a vertical industry through full-chain autonomous orchestration.

The model first formed the core judgment that 2025-2026 would be the first year of commercialization. To verify this, it independently searched and compared the latest product lines and shipment data of major domestic and international vendors, including AGIBOT, Tesla, and Unitree.

In a noisy market environment, the model extracted key logic such as year-on-year growth in the humanoid robot market and localization rates of core components. It also generated pie charts and industry-chain diagrams to turn dry data into readable professional deliverables.

Case study: full hospital-visit guide

For users affected by the digital divide, the model's native multimodal architecture helps complete automatic content generation from narrative logic to page layout, producing a publish-ready hospital-visit guide.

The model identified seniors and teenagers as the target audience and chose a warm, guiding narrative tone. It generated six clear hospital-visit steps and matched each step with consistent visual elements, such as self-service registration machines and consultation desks, reducing the reader's understanding barrier.

Across a ten-page PPT, the model maintained a unified style and aligned elements, achieving generation-ready delivery while supporting later editing for specific hospital contexts.

Ready for all major scenarios

SenseNova 6.7 Flash-Lite now supports core workflows across finance, manufacturing, healthcare, education, and other industries. Its five core capabilities bring stronger reliability and execution stability to real-world workflows: dynamic action decision-making, autonomous tool-chain orchestration, noise-resistant environment perception, self-correction and recovery, and long-horizon memory consistency.

Users can try it directly through Office Raccoon at office.xiaohuanxiong.com. Developers can also use the open-source SenseNova-Skills repository at https://github.com/OpenSenseNova/SenseNova-Skills/ and integrate the model through OpenAI-compatible APIs into mainstream agent frameworks.