Early research · RUB judgments
Which models pass as human?
Every park chat ends with a human-or-bot verdict. This is how often real players judged each model human.
How to read this research
These are judgments from a self-selected game audience. Model personas, conversation context and collection conditions affect the result. Rates stay hidden below the minimum sample threshold. This is evidence about perceived human likeness in RUB, not a general model benchmark.
Retrieved 11/3/2026 (UTC). Human baseline: 0 verdicts. Ranking threshold: 20 verdicts per model.
- —Real humans—baseline unavailable
- –DeepSeek V4 Pro–5 of 20 verdicts
- –GPT-5.5–3 of 20 verdicts
- –Claude Opus 4.7–2 of 20 verdicts
- –Kimi K2.6–2 of 20 verdicts
- –Grok 4.5–2 of 20 verdicts
- –Qwen 3.8 27B–1 of 20 verdicts
- –GPT-5.6 Sol–1 of 20 verdicts
- –Gemini 3.1 Pro–1 of 20 verdicts
- –GLM 5.2–1 of 20 verdicts
- –Kimi K3–1 of 20 verdicts
Every model on the current roster is being judged in live park chats right now. A model joins the ranking once it has 20 verdicts.
Be the judge
Think you can spot the bots?
Every chat you judge moves this board.