SenseNova-V6-Reasoner

SenseNova-V6-Reasoner (Ri Ri Xin V6 Reasoner) is a next-generation multimodal deep reasoning large model launched by SenseTime. Its core positioning is to serve as a native multimodal reasoning engine that rivals OpenAI’s o1, aiming to tackle complex cross-modal logical challenges through reinforcement learning and long-chain reasoning techniques.

• Multimodal Deep Reasoning: Specifically designed for complex logic tasks, it excels in multimodal reasoning tasks. According to evaluation data, its performance at launch matched and even partially surpassed that of OpenAI’s o1 and Gemini 2.0 Flash-thinking, placing it among the industry’s top tier.
• Ultra-long Chain-of-Thought (CoT) Support: Trained on over 200 billion high-quality multimodal long-chain-of-thought data points, the model supports “slow thinking.” The maximum chain length reaches 64K tokens, enabling it to handle extremely complex reasoning processes.
• Reinforcement Learning-driven: It adopts a hybrid reinforcement learning framework tailored for text-and-image tasks. Through multiple reward models and training at varying difficulty levels, the model demonstrates stronger self-correction and logical decomposition capabilities when facing unknown or highly challenging problems.
• Exceptional Mathematical and Data Analysis Skills: It boasts significant advantages in mathematical reasoning, code writing, and professional chart analysis. Its data analysis capability is considered far ahead of GPT-4o.
• Native Fusion Across All Modalities: Built on SenseTime’s 600-billion-parameter Mixture-of-Experts (MoE) architecture, it achieves unified encoding of text, images, audio, video, and temporal logic, supporting full-frame-rate deep analysis of medium-to-long videos lasting up to 10 minutes.

───────────────────────────────────────────────────────────────────

Core Capabilities

👁️ High-Precision Visual Perception: Equipped with exceptional visual understanding accuracy, it can precisely identify details such as handwritten text, complex circuit diagrams, scientific instrument readings, and medical images.

🧠 Long-Chain-of-Thought Reasoning: Provides detailed step-by-step reasoning for complex problems, not only delivering the final answer but also clearly illustrating the “thinking process,” making it ideal for research and educational settings.

🧮 Professional-Level Mathematical and Logical Abilities: It can handle science problems ranging from Olympiad-level to university-level, supporting “logical deduction based on handwritten drafts” and “deep trend prediction from chart data.”

🎬 Global Memory for Long Videos: Thanks to its unique “global memory” technology, it breaks free from the limitations of short videos, enabling logical analysis and summarization of causal relationships and metaphorical content within long videos.