Search “Multimodal”

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

How WorkBuddy Helps Agency Workers Clock Off Early

Tencent WorkBuddy, as an AI workbench, helps agency workers in the advertising and content industries break scattered briefs into communication plans, to-do lists, and PPTs, and also organizes guest schedules and budget sheets. It comes with Skillhub built in, which has already integrated operational skills for platforms such as Xiaohongshu, WeChat Official Accounts, Douyin, and WeChat Channels, and can generate sizes, styles, and copy for different platforms; it also connects to Tencent Miora to maintain a consistent style across multimodal visual assets such as graphics and video, and supports executing tasks in multiple windows.

Ceva 2026 Edge Technology Seminar will be held on November 5 in Zhangjiang, Shanghai

Ceva announced that the 2026 Ceva Edge Technology Seminar will be held on November 5 at the Hilton Hotel in Zhangjiang, Shanghai, targeting semiconductor engineers, SoC architects, and embedded developers. The event is divided into two tracks: AI and Sensing, and Connectivity Technology. Topics cover the memory and bandwidth bottlenecks of LLM inference, NPU architectures for generative and multimodal AI, AI applications below 1TOPS, as well as Bluetooth 7, Wi-Fi 7/8, UWB, and 5G NTN custom chips.

DeepBlue Technology launches Wowoker, the world's first large-model AI agent work phone

DeepBlue Technology has launched Wowoker, the world's first large-model AI agent work phone, aimed at enterprise users such as government and enterprise organizations, state-owned and central enterprises, and financial institutions. Built on Huawei Kirin chips and HarmonyOS, the device integrates multimodal large models, a Decision Model, and the AgentOS agent orchestration system, and supports AI outbound calling, voice-cloned call avatars, AI marketing content generation, and intelligent CRM. Device assets belong to the enterprise, and it supports phone number masking, encrypted retention of communication data in the cloud, and remote locking and data wiping.

Mistral AI Opens the Mistral Large 4 Model Weights: 1T Parameters, 49 Billion Activated Parameters

On October 29 local time, Mistral AI officially opened the model weights of Mistral Large 4 (codename le Chonk). This is a natively multimodal model with 1 trillion total parameters and 49 billion activated parameters, trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own data center in Europe. The company says it scores 48% on the Artificial Analysis Coding Agent Index, roughly on par with GLM-5.3, but consumes 47% fewer Tokens per task on average. The model is released under a modified MIT license: personal, academic, fine-tuning, and conventional commercial use are free, while cloud providers offering MaaS resale need to obtain a separate commercial license; online API pricing is $0.70 per million input Tokens and $2.10 per million output Tokens.

Ant inclusionAI Open-Sources the ZwZ Model Family and the ZoomBench Fine-Grained Perception Benchmark (ICML 2026)

Ant inclusionAI has released the ZwZ model family (4/7/8B) accompanying its ICML 2026 paper, achieving SOTA on multimodal perception benchmarks among open-source models, while requiring only a single forward inference pass and no tool calls at inference time. The method is Region-to-Image Distillation (R2I), which distills high-quality VQA supervision generated from enlarged local regions back into the full image, with 37K training samples. It also introduces ZoomBench, containing 845 VQA items covering 6 fine-grained perception dimensions, on which Qwen2.5-VL-7B achieves only 42.5% average accuracy.

Ant Group's inclusionAI Open-Sources 2B Visual Forensics Model DefakerOne2

Ant Group's GuangJian team has open-sourced DefakerOne2 on GitHub, a 2B-parameter unified visual forensics model built on the Qwen3.5-2B multimodal backbone, covering AI-generated content, DeepFake faces, document tampering, natural image tampering, face and ID anti-counterfeiting, and AI-generated video detection. The model was trained on approximately 50 million samples, employs a three-stage Tide-Style Tuning process from LoRA to full parameters and back to LoRA, and uses entropy-based test-time routing to select between the video and frame pathways. The team claims that it achieves SOTA on 40 of 53 detection benchmarks and on 11 of 14 localization benchmarks, and that the weights and the RRVBench benchmark have been released on ModelScope.

Articul8 Releases Decision Model Bivia, Input Tokens $0.042 per Million

Articul8 Launches Multimodal Decision Model Bivia, Which Can Connect to AI Agents and Choose Not to Proceed When Confidence Is Low. Bivia Weighs Each Decision Step Against the Overall Task Goal, Supports Structured Responses, Can Be Called Midway Through Agent Execution, and Is Trained on Real Production Workflows. Its Input Price Is $0.042 per Million Tokens, with No Subscription Fee.

SenseTime's Lin Dahua: AI Progress Doesn't Depend Only on Stacking Large Models

Lin Dahua, co-founder and chief scientist of SenseTime, said at CITIC Securities' 33rd Investor Forum that the key to AI progress is not just scaling up models, but also smarter ways of building them. He emphasized that the "intelligence density" of data determines learning outcomes, which is why SenseTime developed a native multimodal model that integrates multimodal capabilities into the architecture and training from the very beginning. Lin Dahua expects embodied intelligence to accelerate over the next one to two years, and SenseTime is also continuing to explore world models.

Meta Partners with Five U.S. Universities to Accelerate Medical and Scientific Breakthroughs Using Its Models

Meta is partnering with five U.S. universities—Caltech, Ohio State University, Louisiana State University, Indiana University, and UTHealth Houston—providing $2 million in funding and access to its latest models to support research in medicine, physics, chemistry, biology, computer science, and mathematics using its models. Specific projects include using the open-weight SAM model to monitor tracheostomy tube displacement in infants, using multimodal AI to help dementia patients communicate, and using open-weight models combined with abdominal CT, medical records, and ECG data to identify HFpEF heart failure risk early.

Mistral Large 4 weights go live on Hugging Face; the company says it leads on CyberGym-E2E and…

After a three-week preview, Mistral AI has released the weights of Mistral Large 4 on Hugging Face. This multimodal model was trained on 3,800 Nvidia accelerators and has a parameter count of one trillion, and it is currently still difficult to run on a gaming PC or a single GPU. Mistral says it leads GLM-5.3 and Kimi K3 on the cybersecurity benchmark CyberGym-E2E, and is on par with GLM-5.3 on the agentic coding benchmark Terminal Bench 4.0, but there is no independent evaluation yet; preview pricing is $1.4 per million input tokens and $4.2 per million output tokens.

Diffusion Intelligence releases Clover-35B-A3B-Preview, a 35-billion-parameter diffusion language model

Diffusion Intelligence releases the diffusion language model Clover-35B-A3B-Preview, which uses a MoE architecture with 35 billion total parameters and about 3 billion activated per token, and supports a thinking mode. On the 49 tasks of the web Agent benchmark WindTunnel, its median execution time was 1.52 seconds, lower than the 3.21 seconds of the Jev + Mercury 2.5 combination, with success rates of 98.6% and 95.9% respectively. On code benchmarks, it scored 73.6 on SWE-bench Verified and 51.7 on Terminal-Bench 2.1, and the team said it will subsequently launch a multimodal version and advance on-device adaptation.

Mistral Releases Open Weights for Mistral Large 4 (Le Chonk)

Mistral has released the open weights for Mistral Large 4 (Le Chonk), a 1-trillion-parameter natively multimodal model that continues reinforcement learning on top of the preview version, with improved coding and defensive cybersecurity capabilities. It scores 19.7% on FrontierCode 2.0 and 48% on the Artificial Analysis Coding Agent Index, roughly on par with GLM-5.3, but uses an average of 47% fewer tokens per task; on Terminal-Bench 4.0 it improves by 13 percentage points to 41%. The model supports three reasoning modes, uses a modified MIT-based Mistral Ecosystem Open-Weight license, and is priced via API at $0.70 per million input tokens, $0.07 for cached input, and $2.10 for output.

Ant inclusionAI Open-Sources ZwZ2-4B/9B/27B: A Fine-Grained Vision Model Without Zooming Tools

Ant inclusionAI has open-sourced Zooming without Zooming v2 (ZwZ2), trained on Qwen3.5-4B/9B and Qwen3.6-27B using GRPO reinforcement learning, with training data consisting of 20.4K multi-region VQA samples validated across models. The models are trained on full images and require no zooming tools at inference time, and can be transferred to perception, spatial reasoning, long-context understanding, and multimodal agent tasks. The code, data synthesis, and evaluation pipeline are all released under the Apache 2.0 license.

"Internet Advertising Governance Blue Book in the AIGC Era" Released, Advocating "Using AI to Govern AI"

The "Internet Advertising Governance Blue Book in the AIGC Era" was released on October 24 during the 33rd China International Advertising Festival. It categorizes advertising risks into four types—content violations, technical confrontation, transaction links, and new scenarios—and puts forward governance recommendations from eight aspects. The blue book advocates "using AI to govern AI," employing multimodal large models to simultaneously view images, read text, listen to audio, and consider context to determine violations. Ocean Engine introduced that, relying on its intelligent review system, 90% of materials complete review within as fast as 10 minutes, with efficiency improved by 75%.

Baidu Health Upgrades Wenxin Health Assistant, Launches Haoyi Matrix and Health Camera

Baidu Health announced the upgrade of Wenxin Health Assistant and the launch of the Haoyi Matrix and Health Camera; its AI services have reached over 50 million monthly active users. The Haoyi Matrix includes Haoyi Q&A, Haoyi Appointment, Haoyi Talk, Haoyi Rankings, and Haoyi Reviews. Haoyi Q&A has onboarded over 150,000 real doctors from top-tier hospitals nationwide to provide 24/7 online services, while Haoyi Appointment already covers more than 80% of public top-tier hospitals, offering over 120,000 appointment slots per day. The Health Camera supports typing, dialect voice input, photo capture, and video upload with real-time narration while filming; trained on over 3 million multimodal imaging materials, it can identify the positional relationship between dynamic vital signs and multiple symptom locations.

Moonshot AI releases Kimi K3.1 Pro: 2.8T total parameters, 104B activated parameters, weights scheduled to be opened on November 4

Moonshot AI releases Kimi K3.1 Pro, a multimodal MoE model with 2.8T total parameters, 104B activated parameters, and a 1 million token context window. The Standard and Highspeed versions are now available on the Kimi website, App, Kimi Work, Kimi Code and API. The company says Highspeed offers 2–3x faster output at twice the price of the Standard version. Artificial Analysis ranks it fourth on the Intelligence Index with a score of 55, behind Opus 5.5's 58; on Arena's Code Arena WebDev it ranks 8th with 1692 points, up nine places from K3. Standard pricing is $2 per million tokens for input, $10 for output, and $0.2 for cached input, one-third lower than K3; the weights have not yet been released, and the company promises to open them on November 4.