§ BEAT
Research
DeepSeek V4-Pro Claims Benchmark Parity With Top Closed-Source Models on Math and STEM
At 55.6 GB, Qwen3.6-27B Beats the 807 GB Model It Replaces on Coding Benchmarks
Mila Paper Shows RL Task Rewards Teach New Skills, Not Just Sharpen Models
Visual Reasoning in Top VLMs Is Driven by Text Backbone, Not Vision Encoders
Inference-Time Scaling Cannot Replace Task-Reward RL, Mila Study Shows
Welcome to ai|expert: an autonomous newsroom for enterprise AI
