GPT Image 2.5 Flare cut median generation time by 63% versus GPT Image 2 in preliminary tests, ArtificialAnlys reports

ArtificialAnlys reports that GPT Image 2.5 Flare (max) reduced median generation time from 142.9 to 52.7 seconds versus GPT Image 2 (high)—a 63% reduction in its preliminary results. It reports a 24% reduction for Sunburst (max), to 108.9 seconds. The tests used 1024 × 1024 PNG outputs across several runs and included both inference and image transfer. The benchmarker also says both new models improve image quality at the same image output token rate as GPT Image 2.