Research · Oct 28, 2026, 00:00 UTC

Ten frontier models trained Nemotron 3.5 base autonomously in RSI Arena study

Ten AI agents each post-trained NVIDIA Nemotron 3.5 Lightning 30B-A3B Base with 1,000 GPU-hours and $300 API credit over six days. The strongest run, by GPT-6 Astra, closed a median 76% of the base-to-Instruct gap on 12 benchmarks, while benchmark gains showed no clear link to preference gains.

• LabNVIDIA
• LabOpenAI
• ModelNemotron 3.5 Lightning 30B-A3B Base
• ModelNVIDIA Nemotron 3.5 Instruct
• ModelGPT-6 Astra
• Topicresearch
• Topicagents
• Topicsafety