Grok 4.7 Improves Coding but Uses Twice the Output Tokens

SpaceXAI released Grok 4.7, which scored 56 on Artificial Analysis' Coding Agent Index versus 47 for Grok 4.6 but generates about 81,000 output tokens per task against 38,000. At unchanged rates of $2/$6 per million tokens, measured cost per task rose to $3.74 versus $1.99 for GPT-5.6 Sol.

• Coding Agent Index rose from 47 to 56; DeepSWE from 65% to 73%
• Output tokens per task grew from 38,000 to 81,000
• Cost per task hit $3.74 versus $1.99 for GPT-5.6 Sol
• Token rates unchanged: $2 per million input, $6 per million output