3D destruction physics — tornado, wrecking ball, truss bridge
Reported by the author — we did not run this. What that means
Output
The prompt that built 3D destruction physics
build three self-contained HTML scenes with real physics - A tornado that sucks in a whole field - A wrecking ball taking down an apartment block - An overloaded truck collapsing a truss bridge
Runs
| Model | Harness | Status | Cost | Duration | Output | Trust | Artifact |
|---|---|---|---|---|---|---|---|
| Opus 5claude-opus-5 | — | Completed | $1.40 | not measured | 56k | Reported | — |
| Fable 5fable-5 | — | Completed | $2.82 | not measured | 55k | Reported | — |
| Kimi K3kimi-k3 | — | Completed | $0.55 | not measured | 36k | Reported | — |
| GPT 5.6gpt-5-6 | — | Completed | $0.31 | not measured | 20k | Reported | — |
What this is, and what it proves
Three self-contained HTML scenes, one prompt, four models. The author's verdict: only Opus 5 got all three right — houses climb the funnel and exit the top, the wall breaks where the ball lands and the rubble piles up, the bridge drops the truck into the river. Fable 5 cost twice as much and collapsed its building before the ball touched it. The clips below are the bridge scene.
The prompt behind "3D destruction physics — tornado, wrecking ball, truss bridge" is published in full, and was run by 4 models: Opus 5, Fable 5, Kimi K3 and GPT 5.6. 4 of 4 finished, on figures reported by the author.
- Models
- 4
- Completed
- 4 of 4
- Cheapest completed run
- GPT 5.6 — $0.31
Figures reported by atomic.chat. Not independently verified.
Reported, not measured
Tokens and cost are the author's own figures, published in the post; we did not run this and cannot reconcile them against a harness log. The prompt above is not verbatim — the author published the task line and the three scene descriptions, not the exact string sent to each model, which is why no hash is recorded. The source is a single 2x2 split-screen video playing the three scenes in sequence; each clip here is that video's own quadrant for one model, cropped, over the same 14–22s window so the four are the same moment of the same scene. The per-scene costs burnt into that video ($0.431 + $0.508 + $0.458 for Opus 5, $0.837 + $0.914 + $1.071 for Fable 5, $0.075 + $0.140 + $0.096 for GPT 5.6, $0.169 + $0.196 + $0.185 for Kimi K3) sum to the totals recorded below.