Independent Testing Puts Grok 4.5 Fourth on Intelligence, Hallucination Rate More Than Doubles
Artificial Analysis's own benchmark run ranks xAI's Grok 4.5 fourth on its Intelligence Index at a score of 54, behind Claude Fable 5, GPT-5.5, and Opus 4.8, while its hallucination rate jumped to 54% from 25% for Grok 4.3.
Days after xAI launched Grok 4.5 promising Opus-class capability at a fraction of the price, independent evaluator Artificial Analysis has published its own numbers, and they show a real accuracy-for-confidence tradeoff alongside the intelligence gains. Compare it yourself in the arena.
- Intelligence Index: 54, up 16 points from Grok 4.3's 38, good for fourth place behind Claude Fable 5, GPT-5.5, and Opus 4.8
- AA-Omniscience factuality score: 26, up from 18, as raw accuracy rose from 35% to 52%
- Hallucination rate: 54%, more than double Grok 4.3's 25%
- Coding Agent Index: 76 in Grok Build, on par with GPT-5.5 (Codex)
- Pricing: $2/$6 per million input/output tokens, versus GPT-5.6 Sol's $5/$30
It is a common pattern that larger models know more but are also more confident in their knowledge, and this holds for the latest Grok release. - Artificial Analysis
At roughly 1.5 trillion parameters, about three times the size of Grok 4.3, Grok 4.5 gets more answers right but is also more willing to state wrong ones with confidence, a tradeoff that matters most for use cases where a wrong answer is costlier than no answer.