List your company

Long-horizon game playing on BALROG NetHack

6.8Progression RateGemini-3-Pro
-0.2721.5643.45.236Jul 1, 2026

Evaluation results

MethodLinks
Gemini-3-Procategory=Frontier proprietary (BALROG leaderboard)2026.076.8
Gemini-3.1-Pro-Thinkingcategory=Frontier proprietary (BALROG leaderboard)2026.072.6
Claude-Opus-4.5category=Frontier proprietary (BALROG leaderboard)2026.072
AUTOMEMbackbone=Qwen2.5-32B-Instruct, setup=memory training (loop #2)2026.071.85
Gemini-2.5-Procategory=Frontier proprietary (BALROG leaderboard)2026.071.7
AUTOMEMbackbone=Qwen2.5-32B-Instruct, setup=scaffold opt. (loop #1)2026.071.57
DeepSeek-R1category=Open-weight (BALROG leaderboard)2026.071.4
AUTOMEMbackbone=Qwen2.5-32B-Instruct, setup=memory-as-file-system, v02026.070.42
Qwen2.5-72B-Instructcategory=Open-weight (BALROG leaderboard)2026.070.3
Qwen2.5-7B-Instructcategory=Open-weight (BALROG leaderboard)2026.070
Qwen2.5-32B-Instructcontext-management method=sliding window2026.070
Qwen2.5-32B-Instructcontext-management method=sliding window + chain-of-thought2026.070