Kimi K3 in OpenCode 1.18.32
skills on · OpenCode 1.18.32 + default skills kitKit: default (46 web/Three.js/QA + 33 video) · 79 skills installed; skill use not recorded
The model's own page, run in a sandboxed frame only when you ask (a page that needs its own origin opens in a new tab on a separate demo site). Phones get a full-screen player capped at 1.5x pixel density (2x on larger screens) so it stays smooth; the original page runs uncapped. It starts in its auto-tour (T switches to play mode on a keyboard).
Wall time 43m 10sCost $2.85 reported
Review
No verdict yet.
The full review
Stopped by us. Stopped by us (box row cap reached (2880s); stopped by the box lane so the run ships as partial); not a model failure.
Scores
Not scored: only runs that finished on their own are scored.
Downloads
Files actually published for this run. Original output, site-made previews and post-run conversions are labelled separately.
The exact prompt
as run, 1,421 characters. Every result for this prompt
Build a single self-contained Three.js scene of a fantasy sailing ship at sea, seen from a cinematic three-quarter view. The ship has a carved dragon figurehead at the bow with an open jaw and horns, a wooden hull with visible planking and a raised stern castle, three masts with full canvas sails that billow and a few pennants, rigging lines, and lanterns along the rail. It sails on animated water with rolling waves, foam at the bow, reflections and a sun path on the surface. Behind it a rocky coast with a small harbour town: stone houses with tiled roofs, a lighthouse, a pier and scattered trees on green hills. Warm late-afternoon lighting with a low sun, long shadows, light haze in the distance and a sky with a few clouds. The camera slowly orbits the ship. Everything is built in code from Three.js geometry and materials; no external models or textures. Output ONLY the complete index.html, starting with <!DOCTYPE html> and ending with </html>, with no explanation and no markdown fences. Build it as index.html (plus any local asset files you create) in the current directory. Three.js may load from a pinned CDN. You may serve it locally and open it in headless chromium (/usr/bin/chromium, software WebGL, no GPU) to screenshot and inspect your own output, and iterate on your own until it looks genuinely cinematic before finishing. Close any browser you open as soon as you've taken your screenshot.
How it was made
Versions saved during the run
4 intermediate outputs, in the order the model wrote them.
No transcript was delivered with this result.
Cost breakdown
Headline cost $2.85, as reported by OpenCode (its own bill for the run). The list-price estimate from the token counts is $2.85.
| Tokens by type | Tokens | $ per 1M | Cost |
|---|---|---|---|
| Input (uncached) | 293,755 | $3.00 | $0.8813 |
| Cache write | 0 | $3.00 | $0 |
| Cache read | 2,252,713 | $0.30 | $0.6758 |
| Output | 86,103 | $15.00 | $1.2915 |
| of which reasoning | 59,512 | billed as output |
- Total computed from tokens
- $2.8486
- Total reported by the harness
- $2.8486
- Difference
- matches
OpenCode 1.18.32 reported its own total; token counts x list prices shown beside it. The harness reports reasoning tokens separately from output; here output includes them (26,591 visible + 59,512 reasoning), and both bill at the output rate.
Prices: https://platform.kimi.ai/docs/pricing/chat, 2026-09-24.
Run notes
- frozen prompt (rewritten): threejs_dragon_ship_a_plain.txt: rewritten, original NOT_RECOVERED
- plain OpenCode config copy (providers only; no ECC instructions/agents)
- default skills on: 79 skills in this run's xdg/config/opencode/skills
- opencode export written to a file (not piped)
- Mode: interactive TUI in tmux (160x48), inside bubblewrap sandbox
- Isolation audit: none
- Renderer: three.js r160 WebGL (THREE.WebGLRenderer, ES module via an importmap) loaded from the unpkg CDN, so the page is not fully self-contained; canvas 2D is used only to paint textures (CanvasTexture).
- Card video: none (the run's video.mp4 is not H.264 yuv420p with 30+ frames under 24 MB). Poster: the run's own render.png.
- Kimi K3 served via Fireworks (OpenCode provider fireworks-ai).
- Status: stopped by us. Stopped at the row cap (~50 min) before the harness declared it finished; the page shown is what it had built by then.
- Same frozen prompt as the other rows on this test; configurations differ and all rows are unscored, so this is not a head-to-head result.