Three armies on a voxel battlefield, each commanded by a different local model
October 20, 2026

The RTS Arena

Three armies on a voxel battlefield, each commanded by a different local model

The RTS Arena is out. It is a voxel RTS built first for agent evals: local models command the armies, every order is logged with the reason behind it, and a rated ladder keeps score.

You can also take a side and play it like any RTS. Give orders by voice or with the mouse, and your units handle the small stuff on their own. It plays more like a strategy game than a test of how fast you can click.

A free-for-all between qwen3.8, gemma4 and gpt-oss, with each side’s reasoning for turn 6 down the right
Three local models, one map
The Arena ladder: gemma4:26b-a4b at 1211, gpt-oss:20b and qwen3.5:9b at 894
The ladder so far
A comparison card of the three local models: gpt-oss:20b, qwen3.5:9b and gemma4:26b-a4b
The lineup
Posted to10