• inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.

Model size is still the default shorthand for capability, and this lays out why that shorthand now misleads when picking a model to build on. The GLM-5.3 detail also shows long-horizon RL environments, not scale, driving the current jump in engineering task performance.

• articleGLM-5.3: How Chinese labs keep stride with the frontierNathan Lambert
• articleTraining Compute-Optimal Large Language ModelsJordan Hoffmann et al.
• articleGLM-5.3 achieves 60 on the Artificial Analysis Intelligence Index, on par with KArtificialAnlys

Sign in to comment.

Loading comments…