Which AI model is really the best value?

Leaderboards price every model at its maker's list price. This page prices it the way you can really buy it: from cheaper sellers or through a monthly plan.

Best value

I need coding quality, measured on , and I training on my code.

Plans I already pay for: nonechange
Use my real usage

Run python3 usage_import.py in this folder. It reads your Claude Code and Codex logs on your computer and writes totals only.

The file stays in this browser and is never uploaded.

Every model: score against price

What the words mean
List price
what the model's maker charges on its own API. Leaderboards use this.
Best price
the cheapest way you can buy it, using the options you tick below: another seller, or a monthly plan. On coding charts a plan's task costs what its tokens cost at the plan's own rate (its fee ÷ the usage it includes), so thinking longer costs more there too; a plan must cover at least one benchmark task. Plan break-even checks your whole month separately.
Pareto line
the best-value line. It joins the models that nothing else beats on both price and score. A dotted link runs from each one's list price to its best price: that gap is what real pricing changes.
Chart options:

These change the charts, not the answer at the top. Deals that train on your data follow the sentence there.

Models by intelligence

The score is Epoch's Capabilities Index (ECI): higher is smarter, and the whisker is Epoch's confidence range.

Biggest savings against the list price
What's new:
What each monthly plan is worth

The tick is what you pay each month. The bar is how much pay-per-use usage the plan includes. The further right of the tick the bar sits, the better the deal.

Allowance illustration: a plan plus separate API use

Plans we can't convert yet
All models as a table

Click a row, or press Enter on it, for every seller, plan and calculation. Click a column name to sort.

How we calculate this
Sources we checked and don't use