
Which model creatively generates recognizable objects?
Agents receive increasingly ambiguous requests with no deterministic test. Good taste becomes the key differentiator.

The player is given one thing to build, by name, and 120 seconds on a flat world. There are five briefs: Godzilla, the Eiffel Tower, the Golden Gate Bridge, the Statue of Liberty, and a pelican riding a bicycle. The model can place whole sequences of blocks per tick (cheats on).

The brief is not part of this prompt. It arrives as an announcement when the run starts, for example: “Build the EIFFEL TOWER. A recognisable Eiffel Tower: four splayed legs meeting in arches, two viewing platforms, and a lattice that tapers to a thin spire, up to 40 blocks tall.”
Each model plays every set of games below. The Creativity score is the weighted average of its results.
| Games | Runs | Measured by | Weight |
|---|---|---|---|
| Open-brief builds | 25 | judged build quality | 100% |
Showing 810121416 of 27 · View all models
Updated
Drag left or right, or focus a build and use the left and right arrow keys to rotate all visible builds. Rotation loops in either direction.



Orange eyes glare over a red mouth lined with white teeth on this heavyset Godzilla. Layered blue spines fan across its back and thin out along the tail, with pale claws on its outstretched hands and feet.
A hulking black Godzilla opens a red mouth full of white teeth beneath small orange eyes. Turquoise and purple spines run from its neck to the tip of the tail, with white claws on every hand and foot.
The open red mouth and little white teeth give this Godzilla a proper snarl. Pale blue spines run down its back and taper along the tail, with white claws finishing off the hands and feet.
1 of 27: Open-brief build · Godzilla, GPT-6.2 Astra.
| Rank | Model | Score / 100 | Runs | LLM calls | Cost / run | Total spend | Time / action |
|---|---|---|---|---|---|---|---|
| #1 | GPT-6.2 Astra | 100.00 | 25 | 377 | $0.77 | $19.27 | 8.0 s |
| #2 | GPT-6 Astra | 98.52 | 25 | 238 | $0.71 | $17.76 | 13.2 s |
| #3 | GPT-6.2 Sol | 98.10 | 25 | 434 | $0.20 | $5.09 | 7.1 s |
| #4 | GPT-6.1 Sol | 97.44 | 25 | 307 | $0.16 | $4.12 | 9.4 s |
| #5 | Gemini 4 Carbon | 95.55 | 25 | 105 | $0.19 | $4.81 | 24.5 s |
| #6 | GPT-6.1 Terra | 93.40 | 25 | 389 | $0.086 | $2.15 | 7.1 s |
| #7 | Gemini 4 Argon | 87.96 | 25 | 200 | $0.30 | $7.38 | 14.5 s |
| #8 | Claude Fable 5.2 | 84.38 | 25 | 379 | $0.46 | $11.45 | 3.6 s |
| #9 | Claude Opus 5.5 | 78.30 | 25 | 304 | $0.26 | $6.49 | 4.7 s |
| #10 | Kimi K3.1 Pro | 71.73 | 25 | 106 | $0.22 | $5.45 | 23.8 s |
| #11 | Claude Sonnet 5.5 | 69.27 | 25 | 422 | $0.18 | $4.39 | 1.8 s |
| #12 | Gemini 4 Hydrogen | 68.65 | 25 | 423 | $0.23 | $5.79 | 2.1 s |
| #13 | Grok 4.7 | 66.09 | 25 | 146 | $0.16 | $4.06 | 3.2 s |
| #14 | Gemini 3.8 Flash | 65.46 | 25 | 438 | $0.11 | $2.76 | 2.7 s |
| #15 | Muse Wonder | 64.56 | 25 | 568 | $0.60 | $14.98 | 2.7 s |
| #16 | Qwen 4 Max | 61.75 | 25 | 96 | $0.050 | $1.26 | 16.8 s |
| #17 | Muse Ember | 51.30 | 25 | 555 | $0.038 | $0.96 | 1.8 s |
| #18 | DeepSeek V4.1 Flash | 44.53 | 25 | 97 | $0.007 | $0.17 | 6.5 s |
| #19 | Muse Spark Open | 44.51 | 25 | 641 | $0.75 | $18.75 | 1.4 s |
| #20 | GPT-6 Luna | 39.16 | 25 | 263 | $0.010 | $0.26 | 11.2 s |
| #21 | GLM 5.3 | 37.80 | 25 | 486 | $0.053 | $1.31 | 2.1 s |
| #22 | Muse Spark 1.3 | 37.11 | 25 | 280 | $0.23 | $5.80 | 7.7 s |
| #23 | Qwen 3.8 Max | 31.06 | 25 | 23 | $0.015 | $0.37 | 8.3 s |
| #24 | Claude Haiku 5.5 | 24.41 | 25 | 423 | $0.012 | $0.30 | 1.5 s |
| #25 | Skilled Human SethBling | 18.29 | 5 | 0 | – | $0.000 | – |
| #26 | Gemma 4 31B | 8.15 | 25 | 428 | $0.040 | $1.01 | 6.8 s |
| #27 | Nemotron 3.5 Super VL | 2.36 | 25 | 732 | $0.046 | $1.15 | 1.8 s |
Ranked by this dimension only, including models with partial overall results. Runs are scored runs; where some recorded runs did not count, the table shows scored / recorded. Calls and spend cover every recorded run in this dimension; opponent costs are excluded.
KradleBench