ALEM (Multi-Agent Coordination): leaderboard
ALEM — open-ended multi-agent coordination in language agents: homogeneous teams of three agents cooperate in a survival world, scored by normalised episode return as a percent of the maximum achievable reward (total achievement, medium difficulty).
Metric: Total Achievement % (medium). Source: alem-world.github.io. Saturation forecast: Around 2029. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 18.6 | #58 |
| 2 | GLM-5.3 Flash | 15.9 | #65 |
| 3 | Qwen3.8-Flash-Next | 11.1 | #72 |
| 4 | Gemma 4 31B | 10.1 | #200 |
| 5 | Qwen 3.5 27B | 8.6 | #187 |
| 6 | Qwen 3.6 35B A3B | 8.1 | #226 |
| 7 | GPT-5.4 | 7.4 | #79 |
| 8 | Laguna S 2.1 | 7.4 | #462 |
| 9 | MiniMax-M2.7 | 7 | #258 |
| 10 | Qwen 3.5 122B A10B | 6.9 | #173 |
| 11 | Qwen 3.6 27B | 6.6 | #170 |
| 12 | Gemma 4 26B A4B | 6.5 | #358 |
| 13 | Qwen 3.5 9B | 5.5 | #379 |
| 14 | Qwen 3.5 35B A3B | 5.4 | #256 |
| 15 | Gemma 4 E4B | 2.7 | #767 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=alem-multi-agent-coordination · How It Works · Data refreshed daily, snapshot 2026-10-20.