Know which AI model to use before you start.

Say what you’re about to do — an overnight bulk job, a client deck, a new project. Modelproof checks what you already pay for, the live prices and scores, and names one model and how to use it. Neutral, cost-first, no leaderboard.

Paste one short prompt into Claude Code, Cursor, or any agentic AI — it installs the advisor and asks what you’re about to do.

✓ Every price and score here is sourced and dated. Nothing is guessed — unknowns stay blank.

claude — 96×28
the evidence — check the work

01Compare, side by side

Pick up to three models. Coding score is 0–100 — SWE-bench Verified where published, otherwise a sourced estimate (marked est); the dot + label show confidence. "—" means not publicly sourced. We don't guess.

Browse the full table — all models, sortable

02The model map

Every model placed by what it costs (→) and how capable it is (↑). The glow in the top-left is the value zone — the cheaper-and-stronger corner every model wants to be in. Gold dots are smart buys: nothing is both cheaper and better. Grey dots are beaten on both. Ranked on coding; hover any dot for the numbers.

03What more effort actually buys

The newest models let you turn a dial — effort — that trades money for accuracy on every request. Each line is one model; each dot is one setting of that dial, from low to max. Flat line = you're paying more for nothing. Click a name to isolate it; hover any dot for the step-by-step cost.

04What changed lately

New models, newest first. Add price changes and retirements if you want the whole picture.