By skyforbes 
Hey everyone,
I wanted to share the new Grounded Hallucination Leaderboard from Vectara. Since its release nearly two years ago, our hallucination leaderboard served as the de-facto resource for understanding LLM Hallucinations for #RAG and #Agentic applications, providing a standardized benchmark. This new leaderboard is powered by a benchmark dataset that is larger, more robust, diverse, and significantly more challenging.
Gemini-3 is just added to the leaderboard, and in spite of its stellar performance in other areas, the hallucination rate is quite high: 13.6%
Gemini-2.5 was much better in that sense.