The Nemotron 3.5 Super VL model demonstrates a significant improvement in recalling obscure, niche information compared to other large language models. In a benchmark testing knowledge of a 1998 game's rules, Nemotron 3.5 Super VL made only minor errors, whereas models like GPT-OSS 120B-A5B, Qwen3.5-122B-A12B, and DeepSeek v4 Flash hallucinated key details. The model also performs well on consumer hardware, running at approximately 8-10 tokens per second on dual RTX 5060 Ti GPUs. AI
IMPACT Shows promise for LLMs in specialized knowledge domains, outperforming other models in obscure information retrieval.
RANK_REASON User benchmark of an LLM's performance on niche knowledge recall. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
