AI Model Reviewer

DeepSeek V4.1 Flash in dsh 0.1.5-rc.3

skills off

← Open in the gallerySame prompt, all models

DeepSeek V4.1 Flash's output for Lighthouse on a headland at night

▶ Open live demo

Zoom, inline player and more ways to view
DeepSeek V4.1 Flash's SVG for Lighthouse on a headland at night

Showing a WebP render of the 1.5 MB SVG so the page loads fast; zooming loads the SVG itself.

OKLighthouse on a headland at nightmax effortHome harness

Wall time 13m 40sCost $0.24 estimated8.0/10 · Draft score — AI judge

Open the SVG filePoster render (WebP)

Review

A surprise for $0.24: a dark, atmospheric scene with heavy rain, low mist, a moon with a bright glitter path, and a long beam raking across the rain from a silhouetted lighthouse with lit windows. There's a tiny ship on the horizon.

The full review

A surprise for $0.24: a dark, atmospheric scene with heavy rain, low mist, a moon with a bright glitter path, and a long beam raking across the rain from a silhouetted lighthouse with lit windows. There's a tiny ship on the horizon. The mood is excellent, but the storm part is weak.

Verdict by the AI judge (provisional).

OK. Finished on its own after 13.7 min and produced lighthouse.svg.

Bugs

  • No breaking waves: the sea is calm glitter, not stormy
  • Headland is a vague dark mass; the tower's gallery barely reads
  • Beam and moon compete for the same focal area

Scores

AI judge

8.0/10

Draft score — AI judge

Visitor votes and reviewer grades are not available yet.

CriterionMarks
lighthouse1.5
stormy sea + moonlight1
beam through rain/mist2
atmosphere2
craft1.5
Total (sum of the marks)8 / 10
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. A draft score until the reviewer checks it. Rubric: Wave-1 SVG scoring rubric (AI Model Reviewer).
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

agentic harness (save the file; self-render allowed), 378 characters. Every result for this prompt

Create an SVG of a lighthouse on a rocky headland at night: stormy sea, the lighthouse beam cutting through rain and mist, moonlight on the waves, as atmospheric and painterly as you can. Save it as lighthouse.svg in the current directory. You may render and inspect your own output (headless chromium is available at /usr/bin/chromium) and iterate on your own before finishing.

How it was made

Wall time 13m 40s21 saved versions

Versions saved during the run

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

Version 21 of 21
  1. Version 1 at 1m 41s thumbnailv1 1:41
  2. Version 2 at 3m 03s thumbnailv2 3:03
  3. Version 3 at 4m 14s thumbnailv3 4:14
  4. Version 4 at 5m 06s thumbnailv4 5:06
  5. Version 5 at 5m 47s thumbnailv5 5:47
  6. Version 6 at 6m 29s thumbnailv6 6:29
  7. Version 7 at 7m 21s thumbnailv7 7:21
  8. Version 8 at 7m 42s thumbnailv8 7:42
  9. Version 9 at 8m 04s thumbnailv9 8:04
  10. Version 10 at 8m 35s thumbnailv10 8:35
  11. Version 11 at 9m 17s thumbnailv11 9:17
  12. Version 12 at 9m 38s thumbnailv12 9:38
  13. Version 13 at 10m 00s thumbnailv13 10:00
  14. Version 14 at 10m 21s thumbnailv14 10:21
  15. Version 15 at 10m 53s thumbnailv15 10:53
  16. Version 16 at 11m 04s thumbnailv16 11:04
  17. Version 17 at 11m 36s thumbnailv17 11:36
  18. Version 18 at 11m 47s thumbnailv18 11:47
  19. Version 19 at 12m 19s thumbnailv19 12:19
  20. Version 20 at 12m 30s thumbnailv20 12:30
  21. Version 21 at 13m 02s thumbnailv21 13:02

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 2.6 MB; it is published as a scrubbed, compressed download rather than shown on the page.

Show all 171 steps

Loading the timeline…

Download the full transcript dsh session log, 483 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 $0.24, estimated from the token counts at list price.

Tokens by typeTokens$ per 1MCost
Input (uncached)49,723$0.30$0.0149
Cache write0$0.30$0
Cache read11,171,200$0.01$0.0670
Output128,209$1.20$0.1539
of which reasoning41,605billed as output
Total computed from tokens
$0.2358
Total reported by the harness
none

Tokens summed from the dsh session log per-step usage (inputTokens = uncached); dsh prints no cost.

Prices: api-docs.deepseek.com/quick_start/pricing (read 2026-09-23); off-peak rates, peak (weekdays 01-04 and 06-10 UTC) is 2x, 2026-09-23.

Run notes

one prompt, agentic harness (self-review allowed)wave 1

  • run started in DeepSeek peak hours: 2x rates applied
  • Mode: headless mode (dsh ships no terminal UI; its interactive UI is a web app) — dsh --profile headless, inside bubblewrap
  • Isolation: bwrap: private /tmp, HOME/TMPDIR/DSH_HOME inside run dir; the project files, other runs, /tmp/hx_* not mounted