Free AI Model Nemotron 3.5 Super VL Beats Rivals on Obscure Knowledge

A tech enthusiast recently compared several free AI models to see which one knows the most about very specific, obscure information. They asked about the rules of a 1998 video game called Anno 1602, without allowing the AI to search the web. The models had to rely only on their internal knowledge. Most models got many details wrong, inventing game editions and upgrade requirements. But one model, Nemotron 3.5 Super VL from NVIDIA, stood out. It made only small mistakes, like adding an extra resource requirement, while others were completely inaccurate.

This matters because it shows that free AI models are getting better at handling niche topics. Previously, you might need to pay for a top-tier model to get accurate answers on specialized subjects. Now, a model that runs on your own computer can be surprisingly capable. The user ran Nemotron on two RTX 5060 Ti graphics cards, which are consumer-grade and cost a few hundred dollars each. It generated text at about 8-10 words per second, which is slower than cloud services but still usable.

The catch is that even the best model still made mistakes. It hallucinated a salt requirement for citizens to become merchants and left out a church for settlers. So while it's an improvement, you can't fully trust any AI for critical niche information without double-checking. The user also noted that Gemma 4 31B IT QAT, a smaller model, was surprisingly capable, though it omitted too many details. This suggests that model size isn't everything; training and architecture matter too.

Overall, this test highlights a trend: AI is becoming more accessible and accurate for everyday users. You no longer need a supercomputer to get decent performance on specialized tasks. But always verify important facts, especially when dealing with obscure or detailed information.