interactive tool free · no key · deterministic

Pick the model for the job — with real numbers.

Choose the task and set your ceilings. The pick is a ranked sort over the same dataset that runs /model-race — price per 1M tokens, context window, and benchmark indexes, each stamped with the snapshot's as-of date. No AI call, no invented scores.

The best model is the cheapest one that clears the bar for YOUR task — not the leaderboard winner.

Task lens— sorts by the real benchmark index for the task

Budget— hard ceiling on blended $ per 1M tokens

Constraints— context window + tool calling, from the dataset

> model pick

Task, budget, constraints.

Only filters the dataset can honestly answer are offered. If a dimension is missing from the current snapshot, its filter isn't shown.

Task

Lowest blended token price first (3:1 input:output), quality as tie-break.

Budget · blended $ per 1M tokens (3:1 in:out)

Minimum context window

needs tool callingvolumeM tokens/month (optional — adds a cost estimate)

data as of 2026-07-07 · OpenRouter metadata (no benchmark scores)

Not offered (the dataset can't answer them): Vision / image quality — no vision benchmark in the model-race feed. Writing quality — no writing-specific benchmark; the general lens is the closest honest proxy. Speed filter — no throughput data in the current snapshot.

Dataset: the Boostor model-race snapshot (Artificial Analysis benchmark feed when a key is configured, OpenRouter model metadata otherwise; documented offline snapshot as the last resort), cached up to 12h. See the full board