Benchmark overviewDetailed audit

Luna: Stim vs local toolchain

Select a model, platform, and run to inspect its timing, commands, terminal output, and Settings-screen proof.

Settings-ready comparison

Latest update 2026-09-03 / valid runs only

JavaScript change

Stim reached Settings 73% sooner

Stim2m 8s

451k total tokens$0.024 total cost14 commands

Control7m 54s

2.73M total tokens$0.097 total cost46 commands

Native change

Stim reached Settings 49% sooner

Stim4m 16s

536k total tokens$0.025 total cost15 commands

Control8m 17s

1.89M total tokens$0.065 total cost38 commands

Benchmark machine

Mac mini (Mac16,10)

Apple M4 / 16 GB

System: macOS 26.5.2 (25F84)
Toolchain: Xcode 26.6 (17F113) / Node 26.7.0
Device: iPhone 17 / iOS 26.5

Run audit

Select a command bar to inspect its terminal output.

Settings ready2m 8sprimary outcome

Total tokens451kfull agent run, including cached input

Total cost$0.024full agent run / reported or API-equivalent estimate

Commands14392k cached input

gpt-5.6-luna / javascript / stim

Valid run

Agent turn 2m 34s

Clock starts at the first recorded agent message or shell command. Excludes 6.3s before that activity, including runner startup and any unobserved initial reasoning. Settings proof from dispatch: 2m 14s. Tokens and cost cover the full turn.

What the agent did

Created an isolated worktree, worked on the JavaScript change, and ran Stim's iOS workflow. It reached Settings and captured valid agent-device proof.

Agent / other 1m 29sShell active 1m 5sCommands summed 1m 5sPeak concurrency 1

"Agent / other" is time with no command active; it includes reasoning, tool selection, harness latency, and idle gaps.

0.0s / 2m 34s

Timeline zoom1x

0.0s38s1m 17s1m 55s2m 34s

Agent

Shell

Events

Terminal30s / exit 0

$ stim ios

  device      booting stim-58bd14ac-bench-luna-rc12-gpt-5-6-luna- (iPhone 17 26.5) (<simulator-udid>)
  fingerprint b1232a.. hit (4.3s)
  device      resetting privacy grants and the keychain on <simulator-udid>
  device      stim-58bd14ac-bench-luna-rc12-gpt-5-6-luna- (iPhone 17 26.5) (<simulator-udid-prefix>) adopted (13.1s)
  install     unchanged (stim-58bd14ac-bench-luna-rc12-gpt-5-6-luna- (iPhone 17 26.5) already has this build) (593ms)
  install     dev client prepared (830ms)
  launch      com.appandflow.trailhead (559ms)
  verify      bundle loaded, process alive, stable for 3s -- the first screen may still be rendering (14.1s total)
OK: com.appandflow.trailhead on stim-58bd14ac-bench-luna-rc12-gpt-5-6-luna- (iPhone 17 26.5) (<simulator-udid-prefix>), Metro port 8082 (from cache, 30.2s)
  device      stim-58bd14ac-bench-luna-rc12-gpt-5-6-luna- (iPhone 17 26.5) (<simulator-udid>)
  app         com.appandflow.trailhead
  metro       running on port 8082
  cache       from cache
  compilation cache not run; artifact cache supplied the app
  logs        results/luna-rc12/javascript-stim/stim-home/workspaces/bench-luna-rc12-gpt-5-6-luna-javascript-stim-178--c66d4b812513ab06/logs

Commands in order (14)

Validated proof

Settings screen

Captured by agent-device after it found "Keep saved trail maps available offline". This screenshot completion is the timing endpoint used above.