RTSGameBench - Time-Constrained Production: leaderboard
RTSGameBench (Kim et al., 2026) resource-management mini-game in Beyond All Reason: produce a specified unit composition before a deadline while fending off enemy raids, played through the RTSGameAgent harness (FSM group management plus agentic memory, the same VLM for memory and decisions). Win rate over 30 runs, in percent; higher is better.
Metric: Win Rate (%; 30 runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 100 | #244 |
| 2 | Gemini 3 Flash | 100 | #159 |
| 3 | Qwen 3.5 397B A17B | 100 | #242 |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 100 | #451 |
| 5 | Kimi K2.5 | 97 | #252 |
| 6 | GPT-5.2 | 93 | #172 |
| 7 | Grok 4.1 Fast | 50 | #426 |
| 8 | GPT-5 Mini | 33 | #349 |
| 9 | Qwen 3 VL 235B A22B Instruct | 30 | #521 |
| 10 | Mistral Large 3 | 0 | #819 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rtsgamebench-time-constrained-production · How It Works · Data refreshed daily, snapshot 2026-11-12.