AnalysisAI ModelsJuly 13, 2026
Krea 2 Turbo speed benchmark: 8 model formats on RTX PRO 6000

Fastest at 1024x1024 is int8_convrot at 2.424 s median, while int4_convrot is best for speed/VRAM at 20.6 GiB. At 2048x2048, int4_convrot leads at 12.678 s median, 1.25x faster than BF16.
1 source
More stories today
- Tool automates complex projects with AI agent swarms
- Airtap launches AI agent for iMessage and RCS
- Sam Altman says 'we are now in the singularity'
- AI Mania Is Eviscerating Global Decision-Making
- Rauch: Software factory is product, agents are key