Home / AI News / details

Gemini 3.1-Pro-Preview Tops SuperCLUE-VLM Rankings

2026-03-09 Update Gemini,Rankings,Performance

Google's Gemini-3.1-Pro-Preview has achieved top ranking in the SuperCLUE-VLM February 2026 benchmark with a score of 84.87 points, demonstrating continued leadership in multimodal AI capabilities. The result represents a significant achievement in visual language model performance.

However, the benchmark also reveals rapidly narrowing gaps with Chinese models from ByteDance, Alibaba, and Moonshot AI, which achieved competitive scores and surpassed international competitors in specific visual reasoning tasks. The close competition indicates accelerating global AI development.

The benchmark results highlight the dynamic nature of AI capability development, with multiple players achieving state-of-the-art performance across different dimensions. The update provides valuable comparative data for organizations evaluating multimodal AI solutions.[citation:3]

Share：

NVIDIA Launches Enterprise AI Agent Platform NemoClaw

NVIDIA CEO Jensen Huang announces NemoClaw, an enterprise-grade AI agent platform built on OpenClaw, urging every company to develop an "OpenClaw strategy

Building Custom AI Agents: A Zero-Code Approach

A tutorial on building custom AI agents without writing code, part of Jiangnan University's AI Literacy Training Camp series. The guide covers visual programming interfaces, pre-built templates, and deployment options for non-technical users.[citation:2]

Getting Started with Gemini 3 Canvas for Interactive Content Creation

Tutorial on using Google's Gemini 3 Canvas for creating interactive web games, mini management systems, and dynamic presentations without coding. The guide covers canvas setup, component library usage, and publishing workflows for interactive content.[citation:2]

NotebookLM Deep Dive: Transforming Knowledge Work

Comprehensive review of Google's NotebookLM, testing its document analysis, information extraction, and visualization capabilities across various use cases. The evaluation covers performance with different document types and practical applications for research and analysis.[citation:2]

AI Model 'Poisoning' Exposed: Pay-to-Rank GEO Service Emerges as New Industry

CCTV's 3.15 Gala exposed a new 'GEO' (Generative Engine Optimization) service that allows companies to pay for top rankings in AI model responses by flooding the internet with promotional content, effectively 'poisoning' training data. This emerging industry raises serious concerns about AI reliability and information integrity.[citation:1][citation:5]

a16z Releases Top 100 Generative AI Consumer Apps List: Chinese Apps Dominate Half the Spots

Silicon Valley venture capital firm a16z released its sixth edition of the 'Top 100 Generative AI Consumer Applications' list, revealing that Chinese applications now occupy nearly half of the mobile rankings. CapCut, DeepSeek, Doubao, and Meitu products all rank prominently, while DeepSeek emerges as the only app truly crossing global AI ecosystems.[citation:3]

US Commerce Department Withdraws Proposed Rule on AI Chip Exports

The US Commerce Department has withdrawn a proposed rule regarding AI chip exports, according to government website updates. The move comes amid ongoing global discussions about AI technology regulation and semiconductor supply chain dynamics.[citation:4]

EU Council Moves to Simplify AI Regulatory Rules, Bans AI for Pornography Generation

The European Union Council is advancing efforts to streamline AI regulatory rules while explicitly prohibiting the use of AI for generating pornographic content. The move represents a balanced approach to fostering innovation while addressing ethical concerns.[citation:4]