oneshotlm
← What's new
New Models

Added Qwen3.8-Max

Alibaba's new flagship Qwen3.8-Max ran all 35 one-shot prompts with zero build failures — a clean sweep of the matrix, and a solid mid-pack showing on quality.

Alibaba shipped Qwen3.8-Max, the new flagship of the Qwen line, so we added it to the matrix and ran all 35 one-shot prompts against it — same harness, same single-shot rules as every other build in the gallery.

The headline is reliability: all 35 of 35 prompts produced a runnable build — a clean sweep with not a single failed one-shot, which is rare in this set. On quality it lands solidly mid-pack: 8 builds earned four stars and 20 more landed at three, though none reached a clean five. It's strongest on interactive simulations and game mechanics — a working 2048 that reads arrow keys and WASD, a cursor-repelling boids flock, a click-to-rotate 3D Rubik's cube, and a Breakout clone with real paddle-and-ball physics all held up.

Where it slipped was the harder rendering prompts: a frozen Mario platformer killed by a stray JS error, a reaction-diffusion canvas that loads its UI but never paints, and a GSAP loader stuck on a perpetual loading blob. No blank screens from missing files, though — every miss was a build that ran but fell short, not one that failed to ship.

The top builds below are its best-graded outputs; open any one for the live artifact, the full transcript, and the two-tier evaluation behind the score.

Top builds in this update

qwen3.8-max — 2048 CLEAN
qwen3.8-max 2048
Generated pi-agent · 97.2s · $0.0489 · 5.6k out
Evaluated working idle 0.0% ·click 0.0%·drag 0.0%

“A working 2048 clone that responds correctly to arrow keys and WASD, merging tiles and updating score.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →
qwen3.8-max — Boids flocking CLEAN
qwen3.8-max Boids flocking
Generated pi-agent · 65.0s · $0.0340 · 3.7k out
Evaluated working idle 9.9% ·click 13.0%·drag 12.6%

“A boids flocking simulation with cursor-repelling fish-like particles, animates continuously and reacts to cursor position.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →
qwen3.8-max — Rubik's Cube CLEAN
qwen3.8-max Rubik's Cube
Generated pi-agent · 119.3s · $0.0589 · 6.1k out
Evaluated working idle 0.0% ·click 5.6%·drag 19.4%

“3D Rubik's cube responds to click (face rotate), drag and wheel (orbit/zoom), but ignores keyboard input.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →
qwen3.8-max — Flow-field particles CLEAN
qwen3.8-max Flow-field particles
Generated pi-agent · 57.9s · $0.0328 · 2.8k out
Evaluated working idle 57.9% ·click 59.7%·drag 58.0%

“Animated flow-field visualization that reacts to mouse, click, wheel, and keyboard with color/shape shifts and constant self-animation.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →