The original headline is: "SpaceX Releases Grok 4.5: Token-Efficient Model for Coding Workflows"

Short summary

SpaceX released Grok 4.5, solving SWE-Bench Pro tasks in ~16K tokens versus Claude Opus's ~67K—a 4.2× efficiency gain at 40% lower input cost. Priced at $2/$6 per million tokens versus Opus's $5/$25, token efficiency becomes the key metric for agentic coding loops. For Cursor users, it's built-in; for others, calculate your current task's token spend and benchmark against Grok 4.5.

• •Grok 4.5 achieves 4.2× token efficiency on SWE-Bench Pro tasks compared to Claude Opus ($0.032 input cost vs $0.335)
• •Pricing at $2/M input, $6/M output is 40-75% cheaper than competing models on total token spend
• •Model trained alongside Cursor IDE for real agentic loops; test on existing tasks before migrating workflows

Generated with AI, which can make mistakes.