#xai

Jul 8, 2026 · 3 min read

Grok 4.5 loses on benchmarks and might win anyway

Grok 4.5 trails Fable 5 and GPT 5.5 on the hardest coding tests. It is also a quarter of Opus's price and burns 4x fewer tokens, which might matter more.