RTSGameBench - Free-for-All: leaderboard
RTSGameBench (Kim et al., 2026) four-player free-for-all in Beyond All Reason: the vision-language model plays through the RTSGameAgent harness (FSM group management plus agentic memory, the same VLM for memory and decisions) against three Easy built-in AIs on a fixed map. Rank score averaged over 30 games (1st = 100, 2nd = 67, 3rd = 33, 4th = 0); higher is better.
Metric: Rank Score (0-100; 1st=100, 2nd=67, 3rd=33, 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 | 66 | #159 |
| 2 | Claude Sonnet 4.5 | 57 | #244 |
| 3 | Kimi K2.5 | 40 | #252 |
| 4 | GPT-5.2 | 37 | #172 |
| 5 | Qwen 3.5 397B A17B | 30 | #242 |
| 6 | Grok 4.1 Fast | 30 | #426 |
| 7 | Qwen 3 VL 235B A22B (Thinking) | 27 | #451 |
| 8 | GPT-5 Mini | 18 | #349 |
| 9 | Mistral Large 3 | 13 | #819 |
| 10 | Qwen 3 VL 235B A22B Instruct | 7 | #521 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rtsgamebench-free-for-all · How It Works · Data refreshed daily, snapshot 2026-11-12.