Today for AI
HOT RADAR
llmHEAT 26.4°

AI CLUSTERED EVENT · 10/10/2026

Benchmark Shows Ideogram 4.5 and FLUX 3 Maintain Consistency Over 30 Edits While GPT Image 2.5 Drifts Due to Full Re-rendering

1 reports archived1 independent sourcesupdated 10/10/2026, 00:47:24
Synthesis & Latest Updates
1 Sources Cross-Validated

Artificial Analysis benchmarked four frontier image editing models over 30 consecutive edits, finding that Ideogram 4.5 and FLUX 3 utilize local editing strategies to keep over 95% of the original image intact. In contrast, the top-ranked GPT Image 2.5 Sunburst re-renders the entire image with each edit, leaving only about 20% unchanged and causing significant color and detail drift over multiple turns. Nano Banana 2.1 performed intermediately, maintaining local edits but exhibiting gradual global darkening.

LATEST/Artificial Analysis benchmarked four frontier image editing models over 30 consecutive edits, finding that Ideogram 4.5 and FLUX 3 utilize local editing strategies to keep over 95% of the original image intact. In contrast, the top-ranked GPT Image 2.5 Sunburst re-renders the entire image with each edit, leaving only about 20% unchanged and causing significant color and detail drift over multiple turns. Nano Banana 2.1 performed intermediately, maintaining local edits but exhibiting gradual global darkening.

HEAT TRENDHourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 1 reports · Latest first
  1. Artificial Analysis (@ArtificialAnlys)T1·68 pts
    • Ideogram 4.5 and FLUX 3 maintain high visual consistency (>95% pixels unchanged) over 30 rounds via local editing mechanisms.
    • GPT Image 2.5 Sunburst, despite ranking #1 in single edits, suffers from significant visual drift in multi-turn scenarios due to full-image re-rendering.
    • Multi-turn consistency depends on non-destructive local update capabilities rather than just raw generation quality.