AI

LOLPlay: Benchmarking Real-Time Visual Agents in League of Legends

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 04/11/2026, 05:18:24 UTC.

Briefings

LOLPlay is a new benchmark for visual agents in League of Legends, featuring 36 champion-specific tasks. GPT-6 Astra achieved the highest task success rate of 25.0% at low and medium effort, while its subgoal success rate reached 40.3% at medium effort. However, high effort did not improve either metric, suggesting a cost to additional reasoning in real-time control. The study involves three models evaluated across 324 task evaluations over 324 task evaluations of three models at three reasoning-effort settings.

Key points

• LOLPlay is a benchmark in native League of Legends with 36 champion-specific tasks.
• GPT-6 Astra achieved the highest task success rate of 25.0% at low and medium effort.

Context

This research evaluates visual agents in League of Legends using screenshots and mouse-keyboard actions. It addresses the challenge of coordinating reasoning and action while the environment evolves during inference, providing a structured environment for testing agent capabilities.

Limitations

The study covers 324 task evaluations across three models at three reasoning-effort settings. High effort did not improve metrics, suggesting a cost to additional reasoning in real-time control. The evidence is from an arXiv preprint and does not include results from other models or specific city locations.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 04/11/2026, 05:58:03 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeieuiqmnx3fk34b6zmf3v2ydori3uey5dzumxsw4b7j4m6it5vnfxu