Technology ❯Artificial Intelligence ❯Model Architecture

Mixture of Experts

Inference Efficiency Resource Optimization Multimodal Models Sparse Models Efficiency Context Length

Alibaba Deprecates Hybrid Mode in Qwen3, Launches Dedicated Instruct and Thinking Models

The Qwen3-2507 update raises benchmark scores by splitting Instruct and Thinking models, offering a 256k-token context window.

Meta Faces Backlash Over Use of Experimental AI Model for Benchmark Testing

Meta Launches Llama 4 AI Models, Introducing Scout and Maverick