
AREX-2 Tested: What Happens When the Agent Loop Lives in the Model
BAAI's AREX-2 is a 27B open-weights model trained to try, check and retry by itself for hours. We tested it to see what that means for the agent harness you already have.

BAAI's AREX-2 is a 27B open-weights model trained to try, check and retry by itself for hours. We tested it to see what that means for the agent harness you already have.

NVIDIA's Dory1.0-4B reruns one block of layers for each token. You choose how many times at deploy time, and the weights in memory never change.

Open models like GLM 5.2 now handle most everyday AI work at a fraction of frontier prices. Swapping one in, though, is an engineering project in its own right, and that is where the savings are won or lost.

A haul truck that stops costs thousands of dollars an hour. Here is how digital twin development, remote operation and industrial VR training turn that downtime into measurable ROI for mining, logistics and manufacturing.

Small vision models, fine-tuned for specific jobs, can now run on a phone with accuracy that used to need a server. Here is what that changes for image-driven apps.

Three ways to make an AI feature work, three very different price tags. Here is the decision framework we use to pick the right one, written for the person signing off on the budget.