Artificial Analysis: Claude Fable 5.2 leads DeepMind’s MKA benchmark, at over five times Gemini 4 Carbon’s cost

According to Artificial Analysis, DeepMind’s new MKA benchmark asks agents to turn visually complex files into deliverables such as reports and spreadsheets. Claude Fable 5.2 scores 57% on weighted “Climb” criteria—the harder rubric checks—versus Gemini 4 Carbon’s 54%. Fable averages $31.97 per task; Carbon averages $6.29. That’s a hefty premium for a three-point lead.