Benchmarks

Real tasks, run several times per setup, graded by running the result. Every attempt, cost, and output, including the losses.

Head to head Oct 26, 2026 2 arms · 2 runs

Who pulls the better espresso?

One prompt, one run each: a cup of espresso you stir with your cursor, where speed is force. Opus 5.5 against Opus 5.5 with the threejs-tsl Workbench.

Opus 5.5 Vanilla 1/1
threejs-tsl pompeii/threejs-tsl on Opus 5.5 1/1

Progressive Oct 23, 2026 3 arms · 3 runs

Jetski Racer: who can keep building?

One jet ski game built in three stages by GPT 6.1 Sol, Opus 5.5, and Opus 5.5 with the threejs-game Workbench. Each stage picks up the code the last one left behind.

GPT 6.1 Sol Vanilla 1/1
Opus 5.5 Vanilla 1/1
threejs-game pompeii/threejs-game on Opus 5.5 1/1

Head to head Oct 22, 2026 3 arms · 9 runs

Who's the real Pinball Wizard?

One pinball prompt, three attempts each: Muse Wonder, Opus 5.5, and Opus 5.5 with the threejs-game Workbench.

Muse Wonder Vanilla 3/3
Opus 5.5 Vanilla 3/3
threejs-game pompeii/threejs-game on Opus 5.5 3/3

Build log Oct 26, 2026

Building Vault with three Workbenches

We built a Doppler-style secrets manager from an empty repo with three Workbenches, then hosted it for our own team, fixing the Workbenches as we went. Here's what it cost, how long it took, and how good the result is.

model cost, every run included $43.70
Workbench runs, 33 of them building features 61
total run time, start to finish 16.9 h

The suite Same model, same prompts, same spend

36 working results with Workbenches, 20 without, across 8 tasks.