Gemini 2.5 pro consistently hallucinates when asked about github links

I don’t think this is an issue with T3 Chat, but I don’t have Gemini Advanced account so I couldn’t check this outside of T3 Chat.

I came across the issue when I was using Gemini to try to solve a particular bug I was seeing with a framework, and I asked it to search the web to see if anyone had reported the bug before. It reported back with a github issue link and said it nearly perfectly matched the bug I was describing. I go to the link, and it’s completely unrelated. Just goes to show how little we can trust these things.

So I tried asking it a few times to summarize a few github links. Every time it comes back with information that is seemingly completely hallucinated. The summary seems believable and is related to the URL, but completely different from what’s actually on the webpage. Anyone else able to reproduce this? The only way I was able to get an accurate summary of a github link was through ChatGPT 4o on chatgpt.com.

I believe these models can do web searches, but aren’t able to directly access a particular URL? If that’s the case, why does it not tell me that, instead of making up a summary based on information contained in the URL text.