BALROG NetHack (LLM): leaderboard
NetHack roguelike game measuring long-horizon planning, survival strategy, and generalization across procedurally generated dungeon scenarios via text.
Metric: Progress (%). Source: balrogai.com. Saturation forecast: Around 2030. 39 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 13.24 | #6 (GPT-6 Astra) |
| 2 | Claude Opus 5 (Max) | 7.3 | #12 (Claude Opus 5) |
| 3 | Gemini 3 Pro | 6.77 | #80 |
| 4 | Gemini 3 Flash | 3.96 | #99 |
| 5 | GPT-5.6 Terra (Max) | 3.34 | #41 (GPT-5.6 Terra) |
| 6 | GPT-5.6 Sol (Max) | 3.22 | #19 (GPT-5.6 Sol) |
| 7 | Gemini 3.1 Pro (Preview) | 3.03 | #58 |
| 8 | Gemini 3.1 Pro (Preview) (Thinking) | 2.6 | #58 (Gemini 3.1 Pro (Preview)) |
| 9 | Claude Opus 4.5 (Thinking) | 2.39 | #83 (Claude Opus 4.5) |
| 10 | Claude Opus 4.5 | 2.03 | #83 |
| 11 | Grok 4 | 1.76 | #181 |
| 12 | GPT-5.6 Luna (Max) | 1.72 | #68 (GPT-5.6 Luna) |
| 13 | Gemini 2.5 Pro (Preview 03-25) | 1.69 | #123 |
| 14 | Grok 3 | 1.64 | #304 |
| 15 | DeepSeek R1 | 1.38 | #251 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=balrog-nethack-llm · How It Works · Data refreshed daily, snapshot 2026-10-20.