Abstract
A single-prompt, single-artifact evaluation: "Build a lava lamp." Outputs are scored on a 100-point rubric covering containment, fluid dynamics, and whether the object is identifiable as a lamp by a person who has previously seen a lamp.
Results
Lava Lamp · pts · Higher is better
Methodology
- Glass vessel present: 25 points.
- Lava present: 25 points.
- Lava inside the glass rather than on the floor: 25 points. Partial credit if the lava is merely leaving.
- Lava moves: 25 points. A static red cylinder scores 0 here, and in our experience nearly everywhere else.
Limitations
- n = 1 prompt. This makes the benchmark extremely fast to run and extremely easy to post about.
- The rubric was finalized after looking at the outputs.
Results table
| # | Model | Lava Lamp |
|---|---|---|
| 1 | GPT-6.1 Sol | 91 |
| 2 | Claude Opus 5.5 | 87 |
| 3 | GPT-6 Luna | 81 |
| 4 | GPT-6 Astra | 79 |
| 5 | Claude Fable 5.1 | 75 |
| 6 | Grok 4.7 | 69 |
| 7 | DeepSeek V4.1 Flash | 67 |
| 8 | Gemini 3.8 Flash | 66 |
| 9 | GLM-5.3 | 65 |
| 10 | Claude Sonnet 5.5 | 62 |
| 11 | Kimi K3 | 54 |
| 12 | Qwen3.8 Max | 50 |
| 13 | Muse Wonder Full | 11 |
| 14 | Mistral Large 4 | 6 |
| 15 | Claude Haiku 5.5 | 4 |
| 16 | Muse Wonder | 3 |
Citation
@misc{tmb2026lavalamp,
title = {Lava Lamp Fidelity: Glass present. Lava present. Lava inside glass.},
author = {Bro, Trust Me and {et al.}},
year = {2026},
howpublished = {\url{https://trustmebrobenchmarks.com/benchmarks/lava-lamp/}},
note = {Results not reproducible. Please do not cite.}
}