GPT-5.6 Luna
via codex-cli 0.144.5model + harness
Run of 18 Jul 2026
2 format checks passed — no check of the result
2 render and interaction checks not run
Evaluation not published
Archived pilot · Historical run
Brand-Board
Visual craftbuilt by Claude Fable 5 via claude-code-cli 2.1.214 model + harness
- Automated checks
- 2 format checks passed — no check of the result2 render and interaction checks not run
- Judge
- Evaluation not published
- Started
- Duration
- 131 s
Artifact
Rendered by agi.jetzt from the model-produced artifactRendered offline in an isolated renderer from the stored file (sha256 653ae33a54c0…). The file itself is not served.
Same brief, other models
Scroll sideways. Each run had its own runtime and limits — compare the artifacts, not a score.
Qwen3.8-27B
via ollama 0.32.15model only
Run of 3 Sept 2026
2 format checks passed — no check of the result
2 render and interaction checks not run
Evaluation not published
Automated checks
Automated checks of the stored HTML file, without running it in a browser. Details are the runner’s German originals. Format and presence checks say nothing about whether the result fulfils the brief.
| Check | Kind | Result |
|---|---|---|
| single_html5_filegenau eine HTML5-Datei | Format | Passed |
| viewport_metaViewport-Meta: 1 (erwartet 1) | Format | Passed |
| no_external_requestsSandbox not available — rendering skipped | Rendering | not run |
| renders_without_console_errorsSandbox not available — rendering skipped | Rendering | not run |
2 render and interaction checks were not run. Recorded reason: “Sandbox not available — rendering skipped”.
Evaluation
- Deterministic score
- No score — only the format was checked, not the result
- Judge
- not published for this run
WhyThe v2 judge record of this run names the tested model as its own judge and lists no criteria. Its score is therefore not shown — nor the combined v2 score, which blends that judge in.
Limits of this run
- 2 render and interaction checks were not run. Whether the artifact works in a browser was not checked automatically.
- No executed check with a shown result evaluates what was built.
- No judge evaluation is published for this run. There is therefore no quality verdict — only the automated checks listed here.
- The settings shown are nominal: this runtime was called without them, so they were not enforced.
- Not fully reproducible; the recorded reason is listed under “Environment”.
Task
Brief (turn 1)
Baue als EINE einzelne HTML-Datei (kein Framework, kein externes Asset, keine externen Fonts) ein vollständiges Brand-Board für Balkonknolle — eine App, mit der Stadtbewohner Hochbeete auf dem Balkon planen und bewässern.
Das Board präsentiert die visuelle Identität übersichtlich auf einer Seite und enthält:
- Die Wortmarke / das Logo, rein typografisch gesetzt (kein Bild-Asset), prominent platziert.
- Eine Farbpalette: mindestens 5 Farbfelder (z. B. Primär, Sekundär, Akzent, helles und dunkles Neutral) — jedes mit sichtbarem Hex-Wert und Rollennamen.
- Eine Typo-Skala: mindestens Display, Headline, Body und Caption mit Beispieltext, Schriftgröße und Schriftgewicht.
- Button-States: mindestens Default, Hover und Disabled eines Primär-Buttons (Hover über :hover tatsächlich sichtbar).
- Eine Beispielkarte (z. B. Produkt- oder Info-Karte), die Farben, Typo und Button im Zusammenspiel zeigt.
Stilvorgabe: Freundlich und organisch: satte Grüntöne (Blattgrün, Moos) mit warmem Creme und einem Beeren-Akzent, abgerundete Formen, große Radien, verspielte aber aufgeräumte Sans-Serif. Buttons weich und einladend mit deutlichem Hover-Hub.
Sauber gerastert, klar beschriftet, wie eine echte Marken-Übersichtsseite eines Studios. Responsive, kein horizontaler Overflow.From the task seed (German original); the actual prompts add output-format instructions and other wording.
Environment
- Runtime
- claude-code-cli 2.1.214
- Execution
- model + harnessThe model ran inside an agent harness. The result describes model + harness, not the model alone.
- Nominal profile
- std-v1 · T=0.2 · seed 7331 · 1 turn · no tools
- Profile applied?
- No — claude-code-cli is called without temperature, seed, token or tool settings; the values are the lab’s nominal profile.
- Reproducibility
- not fully reproducibleReasoning default and model snapshot can change on the provider side; CLI version recorded.
- Run window (runner clock)
- 2026-07-18T18:16:38+02:00 → 2026-07-18T18:18:49+02:00
Provenance
- Source system
- agi.jetzt v2 (earlier lab, imported)
- Source run
- 2026-07-18_fable-5_t04_s2_r3
- Status
- passed the v2 lab’s automated technical gates (“verified”)
- Run manifest (sha256)
- 7a53ba5a10979089b3c83332562e39d29c8797bd878f1aaa997bccdf391c6b73
- Artifact (sha256)
- 653ae33a54c0d70b6d4f3796619e9e164f49b8f78f90dd223b9b114abb955df8
- Checks version
- t04-checks-v2
Model The runner’s model identifier maps to Claude Fable 5 in the agi.jetzt register (verified mapping).
What this is evidence for
Not an AGI measureThis run shows one thing: what a model built as a single HTML file from a precise brief. It says nothing about general intelligence.Checked automatically in this run: 2 format checks passed — no check of the result · 2 render and interaction checks not run.
What the task set is meant to measure
- Artifact creation from a precise brief
- Instruction following
- Static checks of the file (no functional test)
Not in its scope
- General intelligence / AGI
- Agents and tool use
- Multimodal input
- Retrieval
- Conversation
Published and unpublished runs
3 of 13 verified v2 runs of this task are published. 3 of the 5 models with recorded runs have a published run.Not published: Claude Haiku 4.5 (3 runs; 0 of 2 format checks passed); GPT-5.6 Sol (3 runs; 0 of 2 format checks passed).Source: the selection records of the published runs. They count the task across its variants and know only models with at least one published run.