oneshotlm
← What's new
New Models

Added Qwen3.8-2.4T-A95B

Alibaba's 2.4-trillion-parameter Qwen3.8-2.4T-A95B turned in the most consistent run of the batch — 35 of 35 builds shipped, zero blank or crashed, and the tightest quality spread in the set.

Alibaba's Qwen3.8-2.4T-A95B — a 2.4-trillion-parameter mixture-of-experts — went through all 35 one-shot prompts on the same harness and single-shot rules as every other build in the gallery.

It's a reliability story more than a ceiling story. All 35 of 35 prompts produced a runnable build, and the quality spread is unusually tight: 21 landed at four stars, 13 at three, and just a single build fell below three — with zero blank screens and no hard crashes anywhere in the set. No clean five, but almost nothing broken, which is rare here.

It's strongest on interactive mechanics and physics: a brick-breaker with lives and a real win state, a chaotic double pendulum with fading trails, a 16-step drum machine, a working Game of Life, and numbered balls colliding and settling inside a spinning heptagon. Its one real stumble was fluid-simulation, which looks right at rest but washes the canvas out once you drag it.

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

qwen3.8-2.4t-a95b — Arpeggiator pad CLEAN
qwen3.8-2.4t-a95b Arpeggiator pad
Generated pi-agent · 72.3s · $0.0766 · 4.3k out
Evaluated working idle 0.0% ·click 0.0%·drag 1.3%

“Pentatonic pad grid with play/stop, tempo slider, and keyboard-triggered pad highlights show interactive arpeggiator response.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →
qwen3.8-2.4t-a95b — Boids flocking CLEAN
qwen3.8-2.4t-a95b Boids flocking
Generated pi-agent · 56.2s · $0.0533 · 3.6k out
Evaluated working idle 10.9% ·click 15.9%·drag 15.5%

“Boids flock and glow nicely, oriented as triangles, with a mouse-following red repulsion zone visible on click/drag/wheel.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →
qwen3.8-2.4t-a95b — Bouncing balls in a heptagon CLEAN
qwen3.8-2.4t-a95b Bouncing balls in a heptagon
Generated pi-agent · 90.4s · $0.0673 · 5.9k out
Evaluated working idle 8.4% ·click 8.9%·drag 8.4%

“Heptagon rotates, 20 numbered balls drop, collide, and settle at bottom with visible spin; no distinct input response but simulation runs well.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →
qwen3.8-2.4t-a95b — Brick breaker CLEAN
qwen3.8-2.4t-a95b Brick breaker
Generated pi-agent · 64.6s · $0.0662 · 4.2k out
Evaluated working idle 0.1% ·click 1.2%·drag 0.3%

“Fully functional brick-breaker with paddle, ball, bricks, score, and lives, ending in a working game over screen.”— vision grade, anthropic/claude-sonnet-5

Full run + transcript →