Best value-for-money LLM on OpenRouter, refreshed every 3 hrs by canonical/model-compare.
| RANK | 7-DAY | MODEL | QUAL | $IN/M | $OUT/M | DISC | PREC | TIER | CTX | AGE | SCORE | TASK$ |
|---|
Prices shown are what a long session pays: several models bill at a higher rate once the prompt outgrows their cheap short-context tier.
TIER marks how prices vary: a token threshold is the prompt size above which the higher rate applies; a percentage is a scheduled off-peak discount (price shown: the standard rate); SCHED marks a schedule whose discount is under 1%. Hover a cell for details.
PREC (precision) is how many bits the model uses to store its parameters — a technique called quantization. Lower precision makes a model faster and smaller to run, but can slightly reduce answer quality. Best-to-cheapest precision levels: FP32, BF16/FP16, FP8, FP6, INT8, FP4, INT4 (highest to lowest quality).
TASK$ is the estimated median cost (in USD) of running a typical task with this model, helping you budget for the operation.
Models with scheduled discount windows, ranked by the size of the discount.
| MODEL | OFF-PEAK | SCHEDULE | QUAL | $IN/M | $OUT/M | CTX | SCORE |
|---|
SCHEDULE shows when the discounted rate applies — long windows are clipped, hover for the full text. $IN/$OUT are the discounted prices during those windows; the standard rate applies at all other times.
No scheduled off-peak discounts in the current data.
Copy to clipboard:
— |
— |
|
|
Tip: save the curl variant as a shell alias, so every
launch picks up the fresh value:
best.txt always serves the balanced #1, regardless
of the tab shown above.
(OpenCode v2 has no --model for TUI launches.) Add the
plugin to opencode.jsonc (project or global):
{ "$schema": "https://opencode.ai/config.json", "plugins": [ { "package": "model-compare-opencode-plugin", "options": { "source": "auto", "defaultPriority": "balanced" } } ] }
It registers three aliases in the model picker —
model-compare/balanced, /price,
/quality — always resolving to the current best. Point your
default at one, or switch any session with /use-best. For
options, keyboard shortcuts, weights, and pins, see
model-compare-opencode-plugin on GitHub.