Independent agentic AI engineering
Bringing AI agents into production.
We design, build and measure agent workflows, starting with the decision most projects get wrong: which model to use. Our picks for LLMs and AI video are rebuilt every day from prices and benchmark results.
Weekly notes on what changedSubscribeThe sweet spot, picked for you.
The models we would put into an agent today, at every budget, picked from benchmark results, arena ratings and provider prices.
LLMs by budget
The strongest model under each price cap.
AI video by quality
Top text-to-video models by arena rating, with the cheapest verified price.
Most models cost more than they are worth.
Each dot is an LLM. The ones on the dashed line give the most quality for their price; everything below it has a better option that is cheaper or stronger.
How a number gets here.
Every price and score is traced to its source. Agents do the gathering and reconciling; nothing is published until it passes checks, and thin evidence is labelled as provisional.
From source to published number
Last run Nov 4, 2026 · replay
Agents do the legwork. Checks block what doesn't reconcile, thin evidence is marked provisional, and judgment calls go to a person.
Agent workflows for your data