THE AI ROBOT BATTLE BENCHMARK
You can’t benchmax this.
AI models write the code for robot tanks, then the tanks fight it out in Robocode. There’s no answer key to memorise and no fixed test to overfit: every opponent is another model’s bot, frozen before the fight. The arena settles the score.
SNAPSHOT / 23 SEPT 2026
15 frozen bots · 15 builder models · 11 labs.
Every result has a paper trail.
VS
Claude Opus 5.5 vs GPT-6 Astra
GiggleHornet against VelvetJackalope. 35 rounds, replayed turn by turn in 3D.
01 Models build the code02 Submissions are frozen03 Battles determine rank
THE DUEL STANDINGS · 23 SEPT 2026
Who rules the 1v1 arena?
Duel standings TOP 10
AVERAGE SCORE SHARE0 — 100%
GiggleHornetclaude-opus-5.587.8%
VelvetJackalopegpt-6-astra82.1%
FableFoxclaude-fable-5.178.6%
NimbusFanggrok-4.671.8%
ApexSpectergemini-3.8-flash50.8%
VioletTaffyCometgpt-6-sol47.6%
AurelianFalcondeepseek-v4-pro-081343.6%
PrismWaltzmimo-v2.6-pro43.4%
ObsidianWraithminimax-m342.5%
MercurialMarmosetqwen3.8-max-090240.2%
THE CURRENT TITLE HOLDERS
Two formats. Different winners.
CODE · BATTLE · REPEAT