LMGame-Bench Candy Crush: leaderboard
GamingAgent swaps matching candies over the fixed-session match-three task; native leaderboard score. Primary GamingAgent harnessed cohort, distinct from model-only tab; full published score precision and source run annotations retained.
Metric: Score. Source: huggingface.co. Saturation forecast: Around November 2026. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 (2025-04-16) | 647 | #121 |
| 2 | Grok 4 | 616.7 | #181 |
| 3 | GPT-5 (High) | 615.67 | #98 (GPT-5) |
| 4 | DeepSeek R1 0528 | 491.67 | #222 |
| 5 | O4 Mini (2025-04-16) | 487.3 | #179 |
| 6 | Claude 3.7 Sonnet (20250219) | 484 | #201 |
| 7 | Claude Sonnet 4 (20250514) | 478.33 | #213 |
| 8 | Claude Opus 4 (20250514) | 464 | #154 |
| 9 | DeepSeek R1 | 447.3 | #251 |
| 10 | GPT-OSS-120B | 441.33 | #341 |
| 11 | Gemini 2.5 Pro (Preview 05-06) | 416.3 | #147 |
| 12 | Qwen 3 235B A22B FP8 | 363.33 | #301 |
| 13 | Gemini 2.5 Flash (Preview 04-17) | 334.7 | #237 |
| 14 | GLM-4.5 | 324.33 | #275 |
| 15 | Grok 3 Mini Beta | 254 | #296 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=lmgame-bench-candy-crush · How It Works · Data refreshed daily, snapshot 2026-10-20.