RTSGameBench - Fixed-Field Skirmish Free-for-All: leaderboard
RTSGameBench (Kim et al., 2026) opponent-modeling mini-game in Beyond All Reason: three or more players with symmetric fixed forces and no production fight until one survives, so the model must read whom each opponent will target, played through the RTSGameAgent harness (FSM group management plus agentic memory, the same VLM for memory and decisions). Rank score over 30 runs (1st = 100 to 4th = 0); higher is better.
Metric: Rank Score (0-100; 1st=100 to 4th=0; 30 runs). Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 64 | #159 |
| 2 | GPT-5.2 | 62 | #172 |
| 3 | Claude Sonnet 4.5 | 55 | #244 |
| 4 | Qwen 3.5 397B A17B | 55 | #242 |
| 5 | GPT-5 Mini | 50 | #349 |
| 6 | Mistral Large 3 | 37 | #819 |
| 7 | Grok 4.1 Fast | 33 | #426 |
| 8 | Qwen 3 VL 235B A22B (Thinking) | 30 | #451 |
| 9 | Qwen 3 VL 235B A22B Instruct | 22 | #521 |
| 10 | Kimi K2.5 | 20 | #252 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rtsgamebench-fixed-field-skirmish-free-for-all · How It Works · Data refreshed daily, snapshot 2026-11-12.