AI Model Reviewer

GLM-5.3 in Claude Code 2.1.280

skills off

← Open in the gallerySame prompt, all models

Unfinished output
GLM-5.3's SVG for PS5 Pro DualSense controller (max-fidelity SVG)

This is the unfinished output at the moment the run ended (Harness + model mismatch). See the review below.

Harness + model mismatchPS5 Pro DualSense controller (max-fidelity SVG)max effortHome harness

Wall time 31m 06sCost $0.92 estimated

Review

Soft, polished white controller on a dark vignette with a strong contact shadow: a dotted touchpad, a face-button diamond in the right colours and outline style, textured concave sticks, a mute bar with an orange line, a speaker-hole grille and small Create/Options buttons. It misses key parts of the brief: no black central core (the whole body is white), the D-pad is four round black buttons with small arrows rather than arrow-shaped buttons, there is no visible blue light-bar glow, and the PS button is a malformed glyph.

Verdict by the AI judge (provisional).

The full review

Harness + model mismatch. Reclassified 2026-09-24 (published as ok): the run ended on 'API Error: 400 This model does not support image inputs' when GLM-5.3 read its own render (…/render1.png) with Claude Code's Read tool; the artifact was already written but the self-review was cut short (Claude Code x text-only model). Re-run on the text-only relay: ps5_controller__claude-code__glm-5.3__attempt2.

Attempt 2 of 3

Every try of this model in this harness on this prompt, oldest first. A later attempt is the cell's result: attempt 3.

  1. Attempt 1 ended in an infrastructure failure, not a model result9m 33s$0.0022026-09-24
  2. Attempt 2 Harness + model mismatch this run31m 06s$0.922026-09-24
  3. Attempt 3 OK the cell's result34m 37s$1.372026-09-25

Bugs

  • no black core section
  • D-pad drawn as round buttons, not arrow shapes
  • no visible light-bar glow
  • PS logo malformed

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

agentic harness (save the file; self-render allowed), 2,074 characters. Every result for this prompt

Create a single self-contained SVG illustration of the Sony PS5 Pro DualSense wireless controller at MAXIMUM fidelity. This is a showcase piece: take your time to plan the geometry carefully before writing it. Requirements: 1) Accurate DualSense silhouette and proportions: a wide body with the large, wide touchpad spanning the top centre, two flared, rounded grips sweeping down and outward, and a perfectly symmetrical left/right layout. White outer shell with the black central core section. 2) Face buttons on the right in the classic diamond: triangle (top), circle (right), cross (bottom), square (left), each drawn as clean PlayStation symbols in their real colours (green triangle, red circle, blue cross, pink square) in the modern outline style on light buttons, with subtle bevels. 3) D-pad on the left as FOUR separate arrow-shaped directional buttons (up, down, left, right), not a single cross. 4) Two black analog sticks with concave tops and a textured grip ring (fine radial or dotted texture) and shading. 5) The PlayStation logo button centred below the touchpad between the sticks, the small mic mute button with its orange-tinted indicator line just below it, the Create button (left) and Options button (right) flanking the touchpad. 6) The blue light-bar glow running along both side edges of the touchpad, with a soft bloom. Speaker grille holes on the black core. 7) Realistic product-render shading: gradients, specular highlights, soft ambient occlusion, a contact shadow beneath the controller. 8) A clean dark studio background with a subtle radial vignette. Front-on hero shot, centred and filling the frame well. Use viewBox="0 0 1600 1100" with width="1600" height="1100". No external assets, no raster images, no fonts, no scripts. Output ONLY the complete SVG code, starting with <svg and ending with </svg>, with no explanation or markdown fences. Save it as controller.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 31m 06sFirst artifact at 28m 55s1 saved version

Versions saved during the run

1 intermediate output, in the order the model wrote them.

Version 1 of 1
  1. Version 1 at 28m 56s thumbnailv1 28:56

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 machine-format transcript is published as a scrubbed download.

Show all 20 steps

Loading the timeline…

Download the full transcript Claude Code session JSONL, 643 KB. 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 $0.92, estimated from the token counts at list price.

Tokens by typeTokens$ per 1MCost
Input (uncached)138,661$1.40$0.1941
Cache write0$0.00$0
Cache read793,494$0.26$0.2063
Output117,019$4.40$0.5149
of which reasoningnot recordedbilled as output
Total computed from tokens
$0.9153
Total reported by the harness
none

Tokens summed from the Claude Code session JSONL (per message id, incl. subagent files); if Claude Code logged zero usage for this unrecognised model, the /cost screen's per-model usage is used instead. Claude Code tags glm-5p3 as unrecognized_model and its own $ figure has costBasis unknown, so it is NOT used: cost = tokens x Fireworks GLM-5.3 price.

Prices: Fireworks GLM-5.3 serverless price (STATE.md 22:42Z; same as the OpenCode/models.dev catalog entry fireworks-ai/glm-5p3), 2026-09-23.

Run notes

one prompt, agentic harness (self-review allowed)PS5 controller (home harness, phase B)

  • Same controller prompt as the earlier home/OpenCode controller runs (harness_runs/prompt_controller.txt).
  • Mode: interactive TUI in tmux (160x48), inside bubblewrap sandbox
  • Isolation: bwrap: private /tmp, HOME and TMPDIR inside run dir, own CLAUDE_CONFIG_DIR copy; the project files, other runs, /tmp/hx_* not mounted