LetTokenBurnBrickBench: Evaluating Agentic Brick Design
A benchmark of 300 text prompts for agentic LEGO-set design, evaluating physical validity, semantic alignment, and design quality across three complexity settings.
$ npm i -g lettokenburn-cli$ ltb bench info brickbench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-6 Astra | 1297.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GPT-6.1 Sol | 1293.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Claude Opus 5.5 | 1249.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Claude Opus 5 | 1101.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Qwen3.8 Flash | 1048.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GPT-5.6 Sol | 1015.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Gemini 3.8 Flash | 1014.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| DeepSeek V4.1 Flash | 1006.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GPT-5.6 Luna | 898.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Muse Spark 1.3 | 868.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GLM-5.3 Flash | 752.0 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
BrickBench contains 300 prompts across Model, Set, and Alt-Build settings. It uses the BrickAgent environment to enable coding agents to construct and validate LEGO assemblies programmatically.
Scoring includes physical validity (connections, collisions, stability), semantic alignment (VQA), and design quality (ELO from pairwise judgments). Leading agents achieve high validity but lag behind human design quality.