Benchmark Prompts / P05
Stone watchtower: 12 AI models, one prompt
Every Season 3 competitor received this prompt verbatim, at default settings, with one generation and no manual cleanup. Nothing below is cherry-picked — the failures ship too, because a benchmark that hides them isn't one.
12Outputs640 ballots · 8 both-bad
A small medieval stone watchtower with a wooden door, arrow slits, and a conical roof.
Every output for this prompt
Click a tile to load the model in 3D, then switch to Wire or X-ray to read the topology. Vote on these blind →
Claude Opus 5.5Stone watchtower Grok 4.7Stone watchtower GPT-6 SolStone watchtower GPT-6 LunaStone watchtower GPT-6.1 SolStone watchtower 3D-AgentStone watchtower GPT-6 AstraStone watchtower Claude Fable 5.1Stone watchtower Claude Opus 5Stone watchtower GPT-5.6 SolStone watchtower Grok 4.6Stone watchtower Claude Fable 5Stone watchtower
Season 3 blind votes for P05
640 ballots · 8 “both bad”
| Competitor | Record | Wins | Losses |
|---|---|---|---|
| GPT-6 Astra | 92 | 21 | |
| 3D-Agent | 90 | 29 | |
| Claude Opus 5.5 | 74 | 20 | |
| Grok 4.6 | 70 | 58 | |
| GPT-6 Luna | 54 | 47 | |
| GPT-6 Sol | 54 | 19 | |
| GPT-6.1 Sol | 51 | 33 | |
| Grok 4.7 | 51 | 45 | |
| Claude Opus 5 | 40 | 77 | |
| GPT-5.6 Sol | 16 | 79 | |
| Claude Fable 5 | 10 | 85 | |
| Claude Fable 5.1 | 6 | 95 |