MoE Models

Twitter

Facebook

Copy Link

Key Facts

Year1991, 2017

Key ModelsGPT-5, DeepSeek V3, Llama 4, Mixtral, Qwen3

Key PeopleRobert Jacobs, Michael Jordan (1991); Noam Shazeer (2017)

Top News

Moonshot AI releases Kimi K2.5 with Agent Swarm that coordinates 100 parallel AI agents

Chinese AI startup Moonshot AI unveiled Kimi K2.5, an open-source large language model that generates web interfaces from visual inputs and coordinates up to 100 sub-agents working in parallel. The model outperforms proprietary models from OpenAI, Anthropic, and Google on key benchmarks while being trained on 15 trillion tokens.

Technology5 sources

Moonshot AI releases Kimi K2.5 with Agent Swarm that coordinates 100 parallel AI agents

Technology5 sources

Phison aiDAPTIV+ delivers 10X faster AI inference on consumer PCs using NAND flash as memory

Phison showcased aiDAPTIV+ at CES 2026, a software and hardware solution that accelerates AI inference by up to 10 times on consumer PCs. The technology uses NAND flash as memory expansion, enabling systems with just 32GB of DRAM to run AI models requiring 96GB through conventional approaches. Partners including Nvidia, AMD, MSI, and Acer demonstrated the technology, which addresses the memory bottleneck limiting AI deployment on ordinary hardware.

Technology3 sources

Phison aiDAPTIV+ delivers 10X faster AI inference on consumer PCs using NAND flash as memory

Technology3 sources

DeepSeek's Engram tackles AI compute waste with conditional memory breakthrough

DeepSeek unveiled Engram, a conditional memory technique that separates static information retrieval from complex reasoning tasks. The method allows AI models to bypass GPU memory constraints by committing knowledge to system RAM, reducing reliance on expensive high-bandwidth memory while improving performance on long-context queries.

Technology5 sources

DeepSeek's Engram tackles AI compute waste with conditional memory breakthrough

Technology5 sources

Nvidia launches Nemotron 3 open source AI models as Meta steps back from transparency

Nvidia released its Nemotron 3 family of models in three sizes—Nano (30B), Super (100B), and Ultra (500B parameters)—with full training data transparency. The move positions the chip giant as a major force in open source AI just as Meta signals a retreat from openness, leaving Chinese models to dominate the space.

Technology16 sources

Nvidia launches Nemotron 3 open source AI models as Meta steps back from transparency

Technology16 sources

NVIDIA AI Servers Achieve 10x Performance Boost for Mixture of Experts Models from DeepSeek and Moonshot

NVIDIA has shattered performance records with its GB200 Blackwell NVL72 servers, delivering a tenfold efficiency boost for mixture-of-expert AI models including those from China's DeepSeek and Moonshot AI. The breakthrough addresses a critical computing bottleneck in scaling MoE models, using 72 chips with high-bandwidth links and 30TB of fast shared memory to enable expert parallelism at unprecedented levels.

Technology3 sources

NVIDIA AI Servers Achieve 10x Performance Boost for Mixture of Experts Models from DeepSeek and Moonshot

Technology3 sources

Mistral AI launches 10 open-source models to challenge OpenAI and Google with edge computing focus

French AI startup Mistral AI released its Mistral 3 family featuring 10 open-source AI models, including a 675B-parameter frontier model and nine smaller variants designed for edge devices. The release positions smaller, customizable models as superior alternatives to large closed-source systems, with capabilities spanning robotics, drones, and offline environments while operating on single GPUs.

Technology11 sources

Mistral AI launches 10 open-source models to challenge OpenAI and Google with edge computing focus

Technology11 sources

Baidu Unveils Next-Generation AI Chips to Challenge Western Dominance in Chinese Market

Chinese search giant Baidu announces new AI training and inference accelerators, including the M100 and M300 chips, alongside rack-scale systems designed to reduce reliance on Western semiconductors amid escalating US-China tech tensions.

Technology2 sources

Baidu Unveils Next-Generation AI Chips to Challenge Western Dominance in Chinese Market

Technology2 sources

Cursor Secures $2.3B Series D at $29.3B Valuation as AI Coding Tools Market Heats Up

AI coding assistant Cursor raises massive $2.3 billion funding round at $29.3 billion valuation, tripling its value in just five months. The company reports over $1 billion in annualized revenue as competition intensifies in the AI development tools market.

Business11 sources

Cursor Secures $2.3B Series D at $29.3B Valuation as AI Coding Tools Market Heats Up

Business11 sources

Chinese AI Startup Moonshot Releases Open-Source Model That Claims to Outperform GPT-5

Beijing-based Moonshot AI has released Kimi K2 Thinking, an open-source reasoning model that reportedly outperforms OpenAI's GPT-5 and Anthropic's Claude Sonnet 4.5 on key benchmarks, costing only $4.6 million to train compared to billions spent by US competitors.

Technology7 sources

Chinese AI Startup Moonshot Releases Open-Source Model That Claims to Outperform GPT-5

Technology7 sources

Cursor 2.0 Launches with Proprietary Composer Model, Promising 4x Faster AI-Assisted Coding

Anysphere releases Cursor 2.0 featuring Composer, their first in-house coding model that completes tasks in under 30 seconds. The update introduces multi-agent parallel workflows and native testing capabilities for enhanced developer productivity.

Technology4 sources

Cursor 2.0 Launches with Proprietary Composer Model, Promising 4x Faster AI-Assisted Coding

Technology4 sources

MiniMax-M2: The New Open-Source AI Powerhouse Challenging Proprietary Models

MiniMax-M2, a newly released open-source AI model, is making waves in the AI community with its impressive performance in coding and agentic tasks, rivaling proprietary models while offering cost-efficiency and enterprise-friendly licensing.

Technology2 sources

MiniMax-M2: The New Open-Source AI Powerhouse Challenging Proprietary Models

Technology2 sources

DeepSeek-OCR: Revolutionary AI Model Compresses Text into Images, Transforming Language Processing

DeepSeek's new open-source AI model, DeepSeek-OCR, introduces a groundbreaking approach to text processing by converting text into images. This method achieves up to 20x compression while maintaining high accuracy, potentially revolutionizing AI language models' efficiency and capabilities.

Technology5 sources

DeepSeek-OCR: Revolutionary AI Model Compresses Text into Images, Transforming Language Processing

Technology5 sources

Reflection AI Raises $2B to Challenge DeepSeek and Build Open-Source Frontier AI

Ex-DeepMind researchers secure massive funding to create an open-source AI powerhouse in the US, aiming to compete with Chinese rivals and closed-source labs.

Technology7 sources

Reflection AI Raises $2B to Challenge DeepSeek and Build Open-Source Frontier AI

Technology7 sources

Prev 12Next