SwarmBench - Flocking: leaderboard
SwarmBench flocking task score, averaged over five simulations of decentralized language-model agents with local perception and communication.
Metric: Task score (mean of 5 simulations, higher is better). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O4 Mini | 17.8 | #185 |
| 2 | DeepSeek V3 | 15 | #329 |
| 3 | Llama 4 Scout | 14.8 | #648 |
| 4 | GPT-4o | 14.6 | #357 |
| 5 | Llama 3.1 70B Instruct | 13.8 | #567 |
| 6 | Claude 3.7 Sonnet | 13.6 | #259 |
| 7 | GPT-4.1 | 13.4 | #258 |
| 8 | O3 Mini | 13.4 | #282 |
| 9 | GPT-4.1 Mini | 12.2 | #363 |
| 10 | Gemini 2.0 Flash | 10.4 | #351 |
| 11 | DeepSeek R1 | 8.4 | #262 |
| 12 | QwQ-32B | 7 | #422 |
| 13 | Claude 3.5 Haiku | 3.6 | #571 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=swarmbench-flocking · How It Works · Data refreshed daily, snapshot 2026-11-05.