Twenty Questions
Frontier
Multi-turn game where models try to guess a secret word/object by asking strategic yes/no questions within 20 turns.
- Domain
- rl-env
- License
- unknown
- Published
- Sep 2025
Cite
Notes
Only stored in your browser.
Top score 1.29 by Qwen3 30B A3B - 13 models reporting (6 frontier)
Score history
13Top models
13
Related tools
1Implementations, trainers, datasets and scaffolds linked to this eval.
FAQ
- What is Twenty Questions?
- Multi-turn game where models try to guess a secret word/object by asking strategic yes/no questions within 20 turns.
- What is the current top score on Twenty Questions?
- The top reported score is 1.29 by Qwen3 30B A3B, across 13 models reporting (6 from frontier labs).
- How can a model improve its Twenty Questions score?
- Tools linked to Twenty Questions on Sophon include Twenty Questions RL Env (Community) - RL environments, datasets, and scaffolds that target this eval.
- What license is Twenty Questions under?
- Twenty Questions is available under unknown.