Metrics Lab

Model telemetry.

Track model quality, spend, device usage, and failure modes across the same public test runs that power DeckTuner.

Runs285
Total Cost$18.46
Input Tokens175147.6k
Output Tokens4091.6k
Cached Tokens140716.9k
Rank by

Model Rankings

#ModelRunsSuccess
1
gemma4
$0.056/run3% wasted↑147836.5k↓2804.3k⚡122777.9k
194
73%
2
kimi-k2.7-code
$0.411/run0% wasted↑1492.4k↓79.3k⚡890.2k
2
50%
3
kimi-k2.7-code
$0.072/run0% wasted↑22832.2k↓1064.0k⚡15362.4k
78
44%
4
gemini-3-flash-preview
$0.113/run0% wasted↑2986.6k↓144.1k⚡1686.5k
11
18%

Cost vs Success Rate

Bubble size = number of runs · lower cost + higher success = top-left is best

Model Efficiency

Lower is better — cheapest and fewest steps per run

Avg Cost / Run

Avg Steps / Run

SteamOS Version

Success rate per model by SteamOS version

ModelCachyOS
gemma4
73%194 runs
gemini-3-flash-preview
18%11 runs
kimi-k2.7-code
35%68 runs
kimi-k2.7-code
50%2 runs