Mixture-of-Experts
-
tech35B Model Matches Trillion-Parameter Performance By Scaling Task Complexity
Shanghai AI Laboratory's Agents-A1 achieves top-tier benchmark results not by expanding parameters, but by extending task sequence length during training.
-
techMeituan's LongCat-2.0: A 1.6 Trillion Parameter AI Model Trained Without US Chips
Meituan unveils LongCat-2.0, a massive open-source coding AI with a 1M token context window, trained entirely on a domestic Chinese 50,000-card compute cluster.
-
techMoonshot AI Launches Open-Source Kimi K2.7 Code Model with 1 Trillion Parameters
Beijing-based Moonshot AI releases Kimi-K2.7-Code, a 1T parameter open-source model reducing reasoning token usage by 30% for efficient AI-assisted programming.
-
techTrillion-Parameter AI Runs on Budget GPU with Legacy Memory
A Chinese enthusiast runs Moonshot AI's Kimi K2.5 on an RTX 3060 using Intel Optane memory, achieving 4 tokens per second.
-
techTencent Unveils Hy3: The Efficient Chinese AI Model Outperforming Rivals
Tencent's new Hy3 AI model, a 295 billion parameter Mixture-of-Experts architecture, achieves remarkable efficiency and performance gains, particularly in coding and reasoning tasks.