AI Model Reviewer

Claude Opus 5.5 in Claude Code 2.1.280

skills off

← Open in the gallerySame prompt, all models

Harness-config failure (ours)

Our harness configuration killed the run before any output. Not a model failure.

Harness-config failure (ours)Humanoid motion cyclehigh effortNeutral harness

Wall time 15m 22sCost $0.70 estimated

Review

No verdict yet.

The full review

Harness-config failure (ours). Our harness configuration cut the run (API Error: Stream idle timeout - no chunks received); not a model failure.

Attempt 1 of 2

Every try of this model in this harness on this prompt, oldest first. A later attempt is the cell's result: attempt 2.

  1. Attempt 1 Harness-config failure (ours) this run15m 22s$0.702026-09-30
  2. Attempt 2 OK the cell's result4m 08s$0.642026-09-30

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.

  • Scrubbed run transcripthumanoid_motion_cycle__claude-code__opus-5-5-high__attempt1-transcript.jsonl · 199 KB · transcript

The exact prompt

as run, 909 characters. Every result for this prompt

Build a single-page Three.js scene (one index.html file, no build step, three.js from a pinned CDN) with a rigged humanoid character (a procedural skeleton and simple body meshes are fine) that plays walk, run and jump cycles, with on-screen buttons or keys to switch between them and blend smoothly from one to the next. The motion must look human, not spider-like: respect realistic joint limits (knees and elbows bend one way only, no joint rotates past a human range), feet plant on the ground with no sliding while in contact, the pelvis bobs and the torso stays upright with natural counter-rotation of shoulders and hips, and arms swing opposite to the legs. Include a ground grid, an orbit camera and a toggle that shows the skeleton. It must run at 60 fps in a desktop browser and work with touch. Write everything into the working directory and finish with a short README.md describing the controls.

How it was made

Wall time 15m 22s

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 3 steps

Loading the timeline…

Download the full transcript Claude Code session JSONL, 199 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 $0.70, estimated from the token counts at list price.

Tokens by typeTokens$ per 1MCost
Input (uncached)2$4.00$0.0000
Cache write5,968$5.00$0.0298
Cache read24,357$0.20$0.0049
Output33,279$20.00$0.6656
of which reasoning33,058billed as output
Total computed from tokens
$0.7003
Total reported by the harness
none

Computed from harness-reported token counts x list prices.

Prices: Project price table (vendor list prices), 2026-09-22.

Run notes

one prompt, agentic harness (self-review allowed)labelled re-run (of humanoid_motion_cycle__claude-code__opus-5-5-high)

  • frozen prompt (written): prompts/humanoid_motion_cycle.md verbatim (written 2026-09-30 from a public X post GPT-6 Sol human motion looks spider-like)
  • plain wave-1-equivalent cc_config copy (settings.json + .claude.json; no ECC files)
  • Mode: interactive TUI in tmux (160x48), inside bubblewrap sandbox
  • Context setting: 1M ([1m] model name)
  • Isolation audit: none