AI Model Reviewer

DeepSeek V4.1 Flash in OpenCode 1.18.32

skills off

← Open in the gallerySame prompt, all models

Zoom, inline player and more ways to view
Poster frame of DeepSeek V4.1 Flash's page
Open the original page

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).

OKAnt vs gardenermax effortNeutral harness

Wall time 26m 41sCost $0.15 reported

Snapshot 9 of 9 saved during DeepSeek V4.1 Flash's run
  1. Snapshot 1 of 9 saved during DeepSeek V4.1 Flash's run
  2. Snapshot 2 of 9 saved during DeepSeek V4.1 Flash's run
  3. Snapshot 3 of 9 saved during DeepSeek V4.1 Flash's run
  4. Snapshot 4 of 9 saved during DeepSeek V4.1 Flash's run
  5. Snapshot 6 of 9 saved during DeepSeek V4.1 Flash's run
  6. Snapshot 7 of 9 saved during DeepSeek V4.1 Flash's run
  7. Snapshot 8 of 9 saved during DeepSeek V4.1 Flash's run
  8. Snapshot 9 of 9 saved during DeepSeek V4.1 Flash's run
Still frames from a low-frame-rate recording made during the run (under 12 fps, so not shown as video): 9 distinct frames.

Generation timelapse

Poster for generation timelapse

Generation timelapse (the agent's screen during the run, sped up). This is not the finished video. Download video (499 KB).

Review

No verdict yet.

The full review

OK. Finished on its own after 26.7 min.

Scores

AI judge

Not scored

No rubric for this prompt

Visitor votes and reviewer grades are not available yet.

AI judge
AI judge: the test runner (model claude-opus-5-5), scoring by visual review of the 1600 px render against the pre-registered rubric. No AI-judge score exists for this result: there is no rubric for this prompt, so no score is shown and none is made up.
Visitors
Public voting with a blind reveal is coming: you score the output first, then see the other scores.
Reviewer
The reviewer's own grade appears here once it is added.

Scores compare results for the same prompt only. AI-judge boards rank runs within one prompt, equal totals sharing a rank; scores from different prompts are never averaged.

Downloads

Files actually published for this run. Original output, site-made previews and post-run conversions are labelled separately.

The exact prompt

as run, 1,384 characters. Every result for this prompt

Build a small playable 3D game in Three.js: an ant defends its home from a furious giant gardener. The player controls the ant on the ground of a lush garden, seen from a low third-person camera that makes the grass blades, pebbles and flowers feel huge. The gardener towers over the scene and sweeps a humming string trimmer back and forth, chasing the ant and throwing up clippings. The ant fights back with a tiny slingshot: aim with the mouse, fire pebbles at the gardener and dodge the trimmer's swings. Show a health bar for the ant and a large health bar for the gardener at the top of the screen, with hit flashes, a clear win or lose message and a restart. Warm afternoon sunlight falls through the leaves overhead, casting moving dappled light and soft shadows, with dust drifting in the light beams. Keep the controls simple (WASD or arrow keys to move, click to shoot) and show them on screen. Everything is built in code from Three.js geometry and materials; no external models or textures. 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 before finishing. Close any browser you open as soon as you've taken your screenshot.

How it was made

Wall time 26m 41sFirst artifact at 4m 25s4 saved versions

Versions saved during the run

4 intermediate outputs, in the order the model wrote them.

Version 4 of 4
  1. Version 1 at 4m 33s thumbnailv1 4:33
  2. Version 2 at 7m 40s thumbnailv2 7:40
  3. Version 3 at 19m 20s thumbnailv3 19:20
  4. Version 4 at 23m 44s thumbnailv4 23:44

Step by step

The prompt, every agent turn and tool call (tool name and a one-line summary; inputs and outputs are not shown), then the final message. The full transcript is 14.5 MB; it is published as a scrubbed, compressed download rather than shown on the page.

Show all 285 steps

Loading the timeline…

Download the full transcript OpenCode session export, 236 KB, gzip. Scrubbed: anything that looks like a key, token, e-mail address or phone number is replaced, and account details and the test machine's file paths are removed.

Cost breakdown

Headline cost $0.15, as reported by OpenCode (its own bill for the run). The list-price estimate from the token counts is $0.15.

Tokens by typeTokens$ per 1MCost
Input (uncached)47,887$0.15$0.0072
Cache write0$0.15$0
Cache read15,076,992$0.00$0.0452
Output166,803$0.60$0.1001
of which reasoning103,194billed as output
Total computed from tokens
$0.1525
Total reported by the harness
$0.1525
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 (63,609 visible + 103,194 reasoning), and both bill at the output rate.

Prices: https://api-docs.deepseek.com/quick_start/pricing, 2026-09-24.

Run notes

one prompt, agentic harness (self-review allowed)prompt rewritten after the 2026-09-24 box losslabelled re-run (of threejs_ant_vs_gardener__opencode__deepseek-flash)

  • frozen prompt (rewritten): engine/prompts_rewritten/threejs_ant_vs_gardener.txt: written by us from the recorded description (zero results)
  • plain OpenCode config copy (providers only; no ECC instructions/agents)
  • opencode export written to a file (not piped)
  • Mode: interactive TUI in tmux (160x48), inside bubblewrap sandbox
  • Isolation audit: outside paths seen