AI Models Struggle with Basic Fact Checks

The smartest AI models on earth are still failing basic fact checks thirty percent of

the time. Google’s new FACTS benchmark just proved that not a single model,

including GPT-5 or Gemini 3, could crack a 70% accuracy score. It gets even worse

with images, where accuracy tanks to under fifty percent. If you use these tools for serious

work, stop trusting the raw output and start verifying every single detail.