SenseNova Unified Large Model (V6)
multimodal
proprietary
SenseTime's flagship multimodal LLM (600B params) integrates text and visual understanding. It achieves top scores on Ch...
Version: V6
Released: 9m 19d ago on 01/13/2025
Pricing:
- details: Cloud/API access (enterprise licensing)
Repository: N/A
Architecture
- parameters: 600 billion
- context_length: 64,000
- inference_type: Mixture-of-Experts Transformer (multimodal)
- training_data: 200B+ multimodal tokens of text-image-video chain-of-thought data
Capabilities
- Chinese
- English
- Advanced multimodal reasoning
- long-context text and video understanding
- data analysis (charts, tables)
- creative generation
Benchmarks
- SuperCLUE: 68.3 (1st in China)
- OpenCompass: Top score (multimodal evaluation)
Safety
- Complies with Chinese content regulations (likely filters politically sensitive prompts)
- Trained with human feedback and reinforcement learning, optimized for factual correctness
- May inherit biases from Chinese-centric training data
Deployment
- regions: Global (China-focused)
- hosting: SenseTime AI cloud platform
- integrations: SenseCore AI platform, enterprise solutions (e.g. Raccoon apps)
Tags
multimodalLLMChinesereasoningAI-cloud