Grok 4.5 Tool Loops: Tune Effort and Cache to Cut Cost

xAI shipped Grok 4.5 on July 8, 2026, and developers moved fast because the numbers are hard to ignore: $2 per 1M input tokens and $6 per 1M output, roughly half of comparable frontier models, served at about 80 tokens/second with a 500K-token context window. xAI also says it solves engineering tasks in far fewer tokens than rivals (about 4x fewer output tokens per SWE-Bench Pro task), so the real bill is lower than the sticker price suggests.