GPTNT (Synchronous): leaderboard
GPTNT (Parekh et al., TMLR 2026) turns the game Keep Talking and Nobody Explodes into a two-agent test in which a Defuser that sees the bomb and an Expert that reads the manual cooperate only through messages. Success rate in percent over single-module missions (10 for each of 11 module types) played by two copies of the same model, pass@1, with the game clock paused while models generate; higher is better. Thinking is disabled and the temperature is 0.6.
Metric: Success rate (%; self-play single-module missions, 10 missions for each of 11 module types, game clock paused while models generate, pass@1). Source: arxiv.org. Saturation forecast: Around August 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 (Non-reasoning) | 39.1 | #158 (Claude Sonnet 4.6) |
| 2 | GPT-5.2 (2025-12-11) (Non-reasoning) | 38.2 | |
| 3 | Gemini 3 Flash (Preview) (Minimal) | 20.9 | #141 (Gemini 3 Flash (Preview)) |
| 4 | Qwen 3.5 27B (Non-reasoning) | 10.9 | #358 (Qwen 3.5 27B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=gptnt-synchronous · How It Works · Data refreshed daily, snapshot 2026-11-12.