Grok 4.5 lands at $2/$6 — and undercuts Opus 4.8 ~17x

xAI just shipped Grok 4.5 across Grok Build, the API, and the console. Headline numbers: a 500K-token context window, text + image input, and pricing of $2 input / $6 output per million tokens (cached input drops to $0.50).

The real story is cost. Opus 4.8 runs $5/$25, so Grok is ~4x cheaper per token — but xAI claims it also solves the same tasks in 4.2x fewer tokens. Stack those together and you get the ~17x cheaper on real work figure making the rounds. Early benchmarks back the "good enough" pitch: 83.3% on Terminal Bench 2.1 (near GPT 5.5) and competitive SWE Bench Pro scores, though it lags on DeepSWE.

Musk is calling it "Opus-class" and aiming it squarely at coding, agents, legal, and finance — the workloads where token bills actually hurt. It's already live as a first-class model in Cursor.

Why it matters: when a frontier-tier model is an order of magnitude cheaper per task, benchmark gaps stop mattering and price becomes the whole game.