Technology ❯ Artificial Intelligence ❯ Machine Learning

Model Architecture

Mixture-of-Experts Mixture of Experts Encoder-Decoder Models Transformer Models Vision-Language Models Sparse Attention Mechanism ExplicitLM Introspective Diffusion Language Model Granite 4.0 Diffusion Transformer

Nvidia Launches Nemotron 3 Super, an Open 120B Model for Agentic, Long‑Context AI

Independent tests show top‑tier throughput with competitive accuracy, with access already live on major model hubs and select clouds.

RAG Shifts From Vector Similarity to Reasoning-Driven Retrieval

Moonshot AI Releases Open-Source Kimi K2.5 With Visual Coding and Agent Swarm

Qwen3-TTS Launches as Open-Source, Low-Latency Multilingual TTS With 3-Second Voice Cloning

DeepSeek Publishes 'Engram' Conditional Memory for LLMs With Reported Gains Over MoE

DeepSeek Publishes mHC to Stabilize Wider Residual Streams in LLMs

Google Releases T5Gemma 2, a Compact Multimodal Encoder-Decoder With 128K Context

Ant Releases LLaDA2.0, Open-Sourcing a Claimed First 100B Diffusion Language Model

Black Forest Labs Debuts FLUX.2, Advanced Image Models With Open Weights and Managed APIs

Ilya Sutskever Says AI’s Scale-First Phase Has Peaked, Urges a Return to Research

Meituan’s LongCat Team Open-Sources LongCat-Video, a Unified Model for Long-Form Video Generation

DeepSeek-OCR Turns Text Into Images to Slash Context Tokens

DeepSeek Open-Sources OCR System That Compresses LLM Contexts With Visual Tokens

DeepSeek-AI Open-Sources 3B DeepSeek-OCR to Compress Visual Context for Long-Document OCR

Samsung’s Tiny Recursive Model Tops Bigger AIs on Tough Reasoning Benchmarks

Samsung Releases 7M-Parameter Recursive AI That Rivals Larger Models on Reasoning Benchmarks

DeepSeek Releases Open-Source V3.2‑Exp With Sparse Attention, Cuts API Prices as Huawei Offers 0‑Day Support

Google Launches Gemma 3 270M to Enable Hyper-Efficient On-Device AI

OpenAI Reinstates GPT-4o Default as GPT-5 Guardrails Show Gaps

Studies Unveil Scaling Laws and Edge Quantization for Mixture-of-Experts Language Models