Model Picker

Which AI model should do a job? The answer depends on the job, how good "good enough" is, and how many times you'll run it. Pick all three below. The data is real: I wrote 36 tasks that a CMMC compliance assessor does, graded 9 models against 1,027 criteria on Artificial Analysis Optima, and recorded the cost of each answer (October 15, 2026).

Source: the benchmark repository. Scores here average the per-task results file, and a task with no score counts as zero, so a few differ from the Optima report's headline means. Your settings stay in this browser.

Your Job

1. What kind of work?
2. What score is good enough?
3. How many tasks a month?
tasks

The Answer

Score against cost per task for each model. Cost uses a log scale.

All Nine Models

Model Score on your mix Cost per task Cost per month Time per task Good enough?

The Full Optima Report

This is the original run on Artificial Analysis Optima, with every task and grade. Open it in a new tab if the frame below is small or blank.

What This Shows

Price did not buy quality on this test. Leave every box checked and set the bar at 85%: the cheapest model that clears it costs about 1% as much as the most expensive one. Then check only "SSP narrative review," the hardest category, and watch the leader change. A single leaderboard number hides that. The right model depends on the work.

That's how I'd approach any AI rollout. Write the real tasks, grade the answers against a rubric, and choose on score and cost together, one job at a time. Want the method for your own work? Email chris@cmm.dev. More work samples.