# Multi-agent Battle on MAgent Battle n=64 agents, 2M training steps (test)

Source: [Multi-agent Battle on MAgent Battle n=64 agents, 2M training steps (test)](<https://www.sota2.com/research/sota/multi-agent-battle-on-magent-battle-n-64-agents-2m-training-steps-test>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-19T10:12:10.486Z

Evaluation of multi-agent performance in the MAgent Battle environment configured with 64 agents, run after 2 million training steps, evaluated on the test set. Performance is measured using reinforcement learning metrics like Final Return, Peak Return, and Convergence Rate.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Multi-agent Battle |
| Dataset (including any stated split) | MAgent Battle n=64 agents, 2M training steps (test) |
| Metric | Final Return |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 9 of 9 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Final Return](<https://www.sota2.com/research/sota/multi-agent-battle-on-magent-battle-n-64-agents-2m-training-steps-test.md?metric=Final%20Return>): Higher is better; unit: SCALAR.

- [Peak Return](<https://www.sota2.com/research/sota/multi-agent-battle-on-magent-battle-n-64-agents-2m-training-steps-test.md?metric=Peak%20Return>): Higher is better; unit: SCALAR.

- [Convergence Rate](<https://www.sota2.com/research/sota/multi-agent-battle-on-magent-battle-n-64-agents-2m-training-steps-test.md?metric=Convergence%20Rate>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Final Return | Result date | Report paper |
| --- | --- | --- | --- |
| HIBCG | 1.38 | Not recorded | 6a0bee9ddbc836b87d82c492 |
| AIB-only | 0.62 | Not recorded | 6a0bee9ddbc836b87d82c492 |
| HIB-flat | 0.62 | Not recorded | 6a0bee9ddbc836b87d82c492 |
| GACG | 0.46 | Not recorded | 6a0bee9ddbc836b87d82c492 |
| QMIX | 0.43 | Not recorded | 6a0bee9ddbc836b87d82c492 |
| ExpoComm | 0.22 | Not recorded | 6a0bee9ddbc836b87d82c492 |
| MAGI | 0.15 | Not recorded | 6a0bee9ddbc836b87d82c492 |
| CommFormer | 0.11 | Not recorded | 6a0bee9ddbc836b87d82c492 |
| BVME | 0.08 | Not recorded | 6a0bee9ddbc836b87d82c492 |

### HIBCG

Final Return: 1.38. Result ID: 6a0c36bafbc373592c02db75.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

Recorded evaluation configuration: {"agents":"n=64","budget":"2M steps"}. This record does not assert independent reproduction.

### AIB-only

Final Return: 0.62. Result ID: 6a0c36bafbc373592c02db72.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

Recorded evaluation configuration: {"agents":"n=64","budget":"2M steps"}. This record does not assert independent reproduction.

### HIB-flat

Final Return: 0.62. Result ID: 6a0c36bafbc373592c02db74.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

Recorded evaluation configuration: {"agents":"n=64","budget":"2M steps"}. This record does not assert independent reproduction.

### GACG

Final Return: 0.46. Result ID: 6a0c36bafbc373592c02db70.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73271e209131cea5ab8a.md>)

Recorded evaluation configuration: {"agents":"n=64","budget":"2M steps"}. This record does not assert independent reproduction.

### QMIX

Final Return: 0.43. Result ID: 6a0c36bafbc373592c02db6d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64671e209131cea046e9.md>)

Recorded evaluation configuration: {"agents":"n=64","budget":"2M steps"}. This record does not assert independent reproduction.

### ExpoComm

Final Return: 0.22. Result ID: 6a0c36bafbc373592c02db71.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa772e1e209131cea719f9.md>)

Recorded evaluation configuration: {"agents":"n=64","budget":"2M steps"}. This record does not assert independent reproduction.

### MAGI

Final Return: 0.15. Result ID: 6a0c36bafbc373592c02db6e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

Recorded evaluation configuration: {"agents":"n=64","budget":"2M steps"}. This record does not assert independent reproduction.

### CommFormer

Final Return: 0.11. Result ID: 6a0c36bafbc373592c02db6f.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73351e209131cea5b287.md>)

Recorded evaluation configuration: {"agents":"n=64","budget":"2M steps"}. This record does not assert independent reproduction.

### BVME

Final Return: 0.08. Result ID: 6a0c36bafbc373592c02db73.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee9ddbc836b87d82c492.md>)

Recorded evaluation configuration: {"agents":"n=64","budget":"2M steps"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
