Gemini 4 Argon cuts hallucinations in tests but answers fewer questions correctly

Google's Gemini 4 Argon matched GPT-6 Astra's score of 53 on the Artificial Analysis Intelligence Index, the benchmarking firm said. Its hallucination rate on a factual knowledge test was 15%, compared with Astra's 51%, but it answered 50% correctly against Astra's 63%. Argon led a business software workflow test with 78%. Each index task cost $1.99 at launch pricing, compared with Astra's $3.26, despite using about 62,000 output tokens against 27,000. At standard pricing, Argon's task cost rises to $3.98. Google has given Argon only to a small group of cyber defenders. Paid API customers and Google AI Ultra subscribers are next, without an announced date.