Simon Willison's command-line tool for language models can now stack templates

Version 0.33, released 22 August, lets the template flag be repeated so one template carries the model and its options while another carries the prompt — which makes running the same prompt across several models a two-flag job. Embedding commands and methods now accept a key per call rather than sharing model state, and reasoning-capable models served through the OpenAI-style endpoint gained a reasoning-summary setting with auto, concise and detailed values.

What it means Separating "which model, configured how" from "what am I asking" is the cheapest way to keep a prompt comparison honest across vendors.