oneshotlm
← What's new
New Models

Added Gemini 3.7 Flash

Gemini 3.7 Flash completed all 35 builds with a 3.00 average across graded samples, showing solid native canvas physics alongside external library loading issues.

On self-contained canvas implementations, the model turned in solid interactive simulations. The vector Asteroids clone shipped with responsive flight controls, wraparound mechanics, and destruction particles, while the boids flocking simulation delivered smooth animation and cursor repulsion zones. Its spinning heptagon demo also rendered working ball-drop collisions and telemetry telemetry overlays, each earning a 4/5.

The misses came primarily from external library integration and minor state logic. Both the D3 force-directed graph and the p5.js particle flow field landed at 1/5 when missing script references left the canvas completely blank behind intact UI shells. Even when builds worked, minor state bugs appeared, such as a 2048 clone that handled tile movement properly but failed to increment the score counter.

Top builds in this update

gemini-3.7-flash — Asteroids CLEAN
gemini-3.7-flash Asteroids
Generated pi-agent · 178.6s · $0.1086 · 30k out
Evaluated working idle 9.2% ·click 11.0%·drag 10.4%

“Vector Asteroids with menu, ship, wraparound asteroids, and destruction particles; ship rotation/movement responds to arrows/wasd.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →
gemini-3.7-flash — Boids flocking CLEAN
gemini-3.7-flash Boids flocking
Generated pi-agent · 136.6s · $0.0715 · 26k out
Evaluated working idle 12.1% ·click 22.4%·drag 20.5%

“Boids flock with pink/cyan triangles, flee from mouse click/drag/wheel creating a repulsion zone; smooth continuous animation.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →
gemini-3.7-flash — Bouncing balls in a heptagon CLEAN
gemini-3.7-flash Bouncing balls in a heptagon
Generated pi-agent · 325.4s · $0.0879 · 28k out
Evaluated working idle 8.5% ·click 8.5%·drag 8.6%

“Numbered balls drop and pile up in a spinning heptagon with physics, collisions, and telemetry; enter/space paused, wasd/arrows had no effect.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →
gemini-3.7-flash — 2048 CLEAN
gemini-3.7-flash 2048
Generated pi-agent · 176.6s · $0.0867 · 28k out
Evaluated working idle 0.0% ·click 0.0%·drag 0.0%

“2048 clone responds to arrow keys and WASD by sliding/spawning tiles, but score stays at 0 despite tile placement.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →