MiniMax-M2.5 as the default worker model for OpenClaw is seriously awesome. Intelligence Index 42 (same level as GLM-4.7) But Blended only costs $0.53/1M tokens (lowest tier on the chart) 80 tok/s + 1.53s latency + 205k context—perfect for running agent loops, coding, or batch tasks at a great value. Recommended strategy: Use M2.5 for regular tasks, and switch to GPT-5.2/Opus for tough problems or critical steps to balance cost and success rate optimally.