Technology ❯ Artificial Intelligence ❯ Model Architecture

Mixture-of-Experts

Parameter Efficiency Efficiency Gemma 4 Variants Parameter Activation Multimodal Models Efficiency Optimization Parameter Count DeepSeek-V3 Kimi Delta Attention Kimi-K2.7-Code Hybrid Latent MoE DeepSeek V3 Training Techniques Parameter Optimization Hybrid Models Performance Benchmarking Heterogeneous MoE Structure Attention Residuals Kimi K3 Performance Metrics Multi-head Latent Attention

6 ARTICLES

2h ago

Alibaba Ships Qwen3.8-Max, Its 2.4T Mixture‑of‑Experts Model, To Public API

The release brings a one‑million‑token, multimodal flagship to hosted APIs as Alibaba promises open weights next week while key documentation and independent evaluations are still missing.

Stories older than 24h

3 ARTICLES

last mo.

Moonshot Releases Open-Source Kimi‑K2.7‑Code as Low‑Cost Coding Model

3 ARTICLES

2mo ago

Nvidia Unveils Nemotron 3 Ultra, a 550B Open-Weight AI Model