6h ago · 9 min read · The AI industry feels like New Year's today. OpenAI dropped GPT-5.5 in the morning, and DeepSeek V4 went live in the afternoon. Add DeepMind's Vision Banana from a couple of days ago, plus Haoyu Cai's Anuttacon paper on digital humans. This week has ...
Join discussion19h ago · 9 min read · A year after DeepSeek’s R1 shocked Silicon Valley — achieving GPT-4-class performance at a fraction of the training cost — the Chinese AI lab returned on April 24, 2026 with its most ambitious open-source release yet. DeepSeek unveiled preview versio...
Join discussion1d ago · 4 min read · Hello everyone, I'm @xiaoqiangapi, a Chinese teacher who has been teaching Chinese for over a decade. Yes, that's the one who, because of one sentence from a student, forced himself to build an API ga
Join discussion1d ago · 6 min read · ### Model Overview DeepSeek V3 is a Mixture-of-Experts (MoE) model designed for high performance in tasks like coding and mathematics.Llama 3.3 70B is an optimized transformer model that excels in multilingual tasks and instruction following. Model D...
Join discussion
1d ago · 6 min read · Key Highlights Model OverviewLlama 3.2 3B: A lightweight, text-only model designed for low-latency applications, optimized for edge devices with 3.21 billion parameters.DeepSeek V3: A powerful Mixture-of-Experts (MoE) model featuring 671 billion para...
Join discussion
1d ago · 6 min read · Today, following the release of FlashMLA, DeepSeek has launched its second OpenSourceWeek project—DeepEP. As the first open-source EP communication library specifically designed for training and inference of MoE (Mixture-of-Experts) models, DeepEP ma...
Join discussion
1d ago · 5 min read · DeepSeek V4 Complete Guide — 1.6T MoE with 1M Context at 73% Lower Cost DeepSeek V4 dropped on April 24, 2026, and it's the most efficient open-weight model family we've seen. A 1.6-trillion-parameter Mixture-of-Experts architecture that only activat...
Join discussion2d ago · 2 min read · DeepSeek R2 dropped in April 2026 and immediately changed the math on reasoning models. A 32B dense transformer that scores 92.7% on AIME 2025, runs on a single RTX 4090, and costs ~70% less than GPT-5 for reasoning tasks. Key Specs PropertyDeepS...
Join discussion3d ago · 9 min read · A year after DeepSeek’s R1 shocked Silicon Valley — achieving GPT-4-class performance at a fraction of the training cost — the Chinese AI lab returned on April 24, 2026 with its most ambitious open-source release yet. DeepSeek unveiled preview versio...
Join discussion4d ago · 8 min read · This is how I measure API overseas latency (taking my Chinese LLM conversion as an example) : 1.45 seconds initial response, do you have a better way? 🎯 ** are you confused too? ** 1.Is it fast to c
Xxiaoqiangapi commented
