English | 中文
A lightweight multimodal agent model built for real-world workflows.
SenseNova 6.7 Flash-Lite is SenseTime's lightweight multimodal agent model, purpose-built for real-world workflows. With a native multimodal architecture that balances quality and cost, it reliably powers complex long-horizon office tasks such as data analysis, slide deck generation, deep research reports, and infographic creation.
SenseNova 6.7 Flash-Lite — a lightweight multimodal agent model built for real-world workflows.
- Lightweight & efficient, balancing quality, cost, and deployability
- Office-tuned, reliably powering complex long-horizon tasks
- Native multimodal architecture, well suited to real office content
- Better token efficiency, keeping complex tasks affordable
- Benchmarks
- Integrated Office Workflow
- Quick Start
- Using with Agent Frameworks
- Token Plan
- Related Links
SenseNova 6.7 Flash-Lite leads across multiple benchmarks, standing out among models of comparable size on long-horizon tasks, planning, and multimodal understanding.
Taking semiconductor memory market analysis as an example, the model covers the entire pipeline from data insight to industry research to content delivery:
Data Insight → Industry Research → Content Delivery
The agent runs the full "read → think → do → deliver" loop on real office tasks. Below are three representative cases with their deliverables.
Query: Read
汇总.csvand clean / analyze the recent memory chip quote data.
Agent Conclusion
Memory prices have been trending upward overall, with select DRAM and NAND products showing the largest gains. The inflection point appeared in late February, followed by an acceleration through March. Categories diverged sharply — server-grade products outperformed consumer ones — indicating that this rally is not uniform but is being led by key categories.
Query: Based on the data analysis, investigate the main drivers of memory and flash price movements since the start of 2026.
Agent Conclusion
The current rally is driven by a combination of supply contraction, surging AI-server demand, and deliberate output discipline by some manufacturers. In the short term, sentiment and inventory restocking amplify volatility, but over the medium term it looks more like a structural rebalancing of supply and demand. If high-end demand persists and OEMs maintain a cautious supply stance, prices may continue to rise or stay elevated.
Memory Price Research.pdf · Research · Report
Query: Generate a 15–20 page Chinese PPT titled "2026 Memory Price Volatility Analysis & Market Outlook".
Agent Conclusion
The final deck follows a clear narrative: first prove with data that "prices are indeed rising and the gains are concentrated in key categories", then explain "why and what's driving it" via external research, and finally give a forward-looking judgement plus actionable recommendations — focus on high-momentum categories, lock in procurement timing, and continuously track OEM strategy and downstream demand.
Semiconductor Memory Market Surge · PPT · Showcase
Note: The above capabilities must be delivered by the Agent framework together with Skills — calling the model API directly cannot reproduce the full workflow.
- Recommended path: Pair OpenClaw or hermes-agent with the official skill library from OpenSenseNova/SenseNova-Skills (see Using with Agent Frameworks below).
- Self-integration: For other agent frameworks, grab Skills directly from OpenSenseNova/SenseNova-Skills and install them yourself.
- Register and complete identity verification at https://platform.sensenova.cn/console.
- From the console sidebar, go to Management Center → API Key Management → Create API Key, then copy and store it safely (the full key is shown only once on creation).
- Set the environment variable:
export SENSENOVA_API_KEY="your_api_key_here"
curl 'https://token.sensenova.cn/v1/chat/completions' \
-H "Authorization: Bearer $SENSENOVA_API_KEY" \
-H 'Content-Type: application/json' \
-d '{
"model": "sensenova-6.7-flash-lite",
"max_tokens": 2000,
"messages": [{"role": "user", "content": "Hi, please briefly introduce yourself."}]
}'For the full API reference (multi-turn, multimodal input, streaming, OpenAI SDK, error codes, etc.), see API.md.
SenseNova 6.7 Flash-Lite needs an agent runtime + the official skill library to deliver an end-to-end office-task workflow.
- Recommended runtime: OpenClaw or hermes-agent.
- Recommended LLM: pair it with the SenseNova platform API (free token plan available).
- Install & setup: see SenseNova-Skills INSTALL.md.
Recommended: just ask the agent to install them for you. Hand it the repo URL and let it clone and copy the contents into the right directory, e.g.:
"Please install https://github.com/OpenSenseNova/SenseNova-Skills into your skills directory."
After installation you may need to restart the agent service manually before the new skills are picked up.
| Agent | Target directory |
|---|---|
| OpenClaw | ~/.openclaw/skills/ |
| hermes-agent | ~/.hermes/skills/ |
Prefer to install manually?
Clone the repo, then copy (or symlink) the subdirectories under skills/ into the target directory:
git clone https://github.com/OpenSenseNova/SenseNova-Skills.git --depth=1
mkdir -p ~/.openclaw/skills
cp -r SenseNova-Skills/skills/* ~/.openclaw/skills/For Hermes, just swap the directory to ~/.hermes/skills/.
The latest model and full Cowork-Skill suite from this repo are integrated into Raccoon, offering enterprise-grade security plus a smooth out-of-the-box experience — and it's free to use. If you'd rather not set up your own environment or manage API keys, you can access these capabilities directly through Raccoon.
Raccoon ships a comprehensive upgrade across product capabilities and client experience:
- Three core office capabilities upgraded: Powered by SenseNova 6.7 Flash and Cowork-Skill, data analysis, PPT generation, and task planning are all further strengthened — covering the full loop of complex knowledge work, from multi-file cleaning and analysis, to formal reporting decks, to industry research / competitive analysis / investment research reports.
- New infographic generation: Built on the SenseNova U1 model, it compresses complex data, long-form reports, and business insights into high-density, structured, visualized infographics — making complex content easier to understand and easier to share.
- All-new client + local Agent OS: Cloud models handle complex reasoning and multimodal understanding, while the local Agent OS centers on local files, work context, and personal habits — delivering a more personalized, localized, and secure AI-native office experience.
- Validated at scale: Trusted by 15 million individual users and thousands of enterprises.
👉 Try it now: office.xiaohuanxiong.com/home
More than a discount — a productivity capability pack for the office
SenseNova Token Plan isn't only "cheap to use" — it's "comfortable to use". We pair curated, more efficient models with generous quota guarantees, so even long-running tasks can run with confidence — delivering a high-value, high-frequency, scalable productivity capability pack for the office.
01 · Native multimodal agent
A multimodal foundation that bridges understanding and generation — reads documents, images, and tables, and produces high-quality deliverables that combine text and visuals.
02 · Tuned for enterprise office work
Closes the loop from "understanding the task" to "producing the final deliverable" — complex office workflows run end-to-end, not just stopping at advice.
03 · ~60% token savings
Token consumption stays under control on complex tasks, raising deliverable output per unit cost — so long-horizon tasks can "run with confidence".
Have questions about integrating SenseNova 6.7 Flash-Lite, hit a bug while using the Skills, or just want to swap notes on real-world workflows? Scan the QR code below to join our WeCom community — the team is there to provide technical support, gather your feedback, and ship improvements based on what you tell us.
| Discord | WeChat Group |
 |
 |
- Website: https://www.sensenova.cn/
- API documentation: API.md
- Raccoon: office.xiaohuanxiong.com/home