Technology ❯ Artificial Intelligence ❯ Machine Learning

Model Architecture

Mixture-of-Experts Mixture of Experts Transformer Models Gemma 4 Encoder-Decoder Models Diffusion Models Vision-Language Models Multimodal Models Sparse Attention Mechanism World Models

8 ARTICLES

last wk.

Moonshot Publicly Launches Kimi K3, an Open-Weight Model With 1 Million-Token Context

The Beijing startup says K3 uses a Mixture-of-Experts design to cut per-query compute and will publish full model weights by July 27 to let developers run and modify it.

17 ARTICLES

last wk.

Thinking Machines Releases Inkling, a 975B Open-Weight AI Model

3 ARTICLES

3w ago

DeepReinforce Unveils Ornith‑1.0, Open‑Source Models That Learn Their Own Coding Scaffolds

9 ARTICLES

4w ago

Chinese AI Firms Slash Training and Inference Costs With Sparse MoE and H800 Workarounds

4 ARTICLES

last mo.

Z. AI Releases GLM-5.2 With 1 Million‑Token Context and Open MIT Weights

3 ARTICLES

last mo.

Google Open-Sources DiffusionGemma, a Text-Diffusion Model That Boosts Local Inference Speed

6 ARTICLES

last mo.

Apple Rolls Out 20B On‑Device AI and Private Cloud GPU Partnership

9 ARTICLES

last mo.

Google Releases Gemma 4 12B to Run Multimodal AI on 16GB Laptops

4 ARTICLES

last mo.

Researchers Propose Training‑Free Tools That Dramatically Speed Diffusion Language Models

5 ARTICLES

last mo.

JetBrains Open-Sources Mellum2, a 12B Mixture-of-Experts Model

4 ARTICLES

2mo ago

NVIDIA Releases Nemotron‑Labs Diffusion With Three Decoding Modes in One Checkpoint

4 ARTICLES

2mo ago

Gemma 4 Emerges as a Local-First Alternative to Cloud AI

10 ARTICLES

2mo ago

MoE Research Delivers New Playbooks for Design, Routing, and Real-World Deployment

9 ARTICLES

2mo ago

Thinking Machines Introduces Interaction Models for Real-Time, Multimodal AI

3 ARTICLES

2mo ago

DeepSeek Releases Multimodal Model, Pitches 'Visual Primitives' for Spatial Reasoning

6 ARTICLES

4mo ago

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

5 ARTICLES

4mo ago

RAG Shifts From Vector Similarity to Reasoning-Driven Retrieval

7 ARTICLES

5mo ago

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

3 ARTICLES

6mo ago

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

4 ARTICLES

6mo ago

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

Model Architecture

Moonshot Publicly Launches Kimi K3, an Open-Weight Model With 1 Million-Token Context

Thinking Machines Releases Inkling, a 975B Open-Weight AI Model

DeepReinforce Unveils Ornith‑1.0, Open‑Source Models That Learn Their Own Coding Scaffolds

Chinese AI Firms Slash Training and Inference Costs With Sparse MoE and H800 Workarounds

Z. AI Releases GLM-5.2 With 1 Million‑Token Context and Open MIT Weights

Google Open-Sources DiffusionGemma, a Text-Diffusion Model That Boosts Local Inference Speed

Apple Rolls Out 20B On‑Device AI and Private Cloud GPU Partnership

Google Releases Gemma 4 12B to Run Multimodal AI on 16GB Laptops

Researchers Propose Training‑Free Tools That Dramatically Speed Diffusion Language Models

JetBrains Open-Sources Mellum2, a 12B Mixture-of-Experts Model

Never miss stories about

Model Architecture

NVIDIA Releases Nemotron‑Labs Diffusion With Three Decoding Modes in One Checkpoint

Gemma 4 Emerges as a Local-First Alternative to Cloud AI

MoE Research Delivers New Playbooks for Design, Routing, and Real-World Deployment

Thinking Machines Introduces Interaction Models for Real-Time, Multimodal AI

DeepSeek Releases Multimodal Model, Pitches 'Visual Primitives' for Spatial Reasoning

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

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

Never miss stories about

Model Architecture