BalatroBench: leaderboard
BalatroBench compares default-strategy tool-calling agents on full Balatro runs. Average final round and run win rate use all published runs, including incomplete runs. The captured v1.0.8 cohort uses the Default1.0.0 strategy, Red deck and White stake; run counts and benchmark version remain attached to each source observation. Native final-round counters can exceed 24, so no finite upper bound is assumed.
Metric: Average Final Round. Source: balatrobench.com. Saturation forecast: Around February 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 19.33 | #71 |
| 2 | GPT-5.2 | 18.38 | #115 |
| 3 | Gemini 3 Flash (Preview) | 16.73 | #90 |
| 4 | Claude Opus 4.5 | 13.87 | #87 |
| 5 | Claude Sonnet 4.5 | 11.87 | #148 |
| 6 | Grok 4 | 9.47 | #189 |
| 7 | DeepSeek V3.2 | 8.67 | #210 |
| 8 | Grok 4.1 Fast | 7.67 | #227 |
| 9 | Qwen 3 Max | 7.27 | #213 |
| 10 | GPT-OSS-120B | 6.27 | #346 |
| 11 | GPT-OSS-20B | 5.8 | #523 |
| 12 | Kimi K2 (Thinking) | 5.47 | #252 (Kimi K2) |
| 13 | Kimi K2.5 | 5.33 | #152 |
| 14 | Claude Haiku 4.5 | 5.07 | #287 |
| 15 | GLM-4.7 | 3.07 | #202 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=balatrobench · How It Works · Data refreshed daily, snapshot 2026-11-05.