|
SpaceX AI shipped Grok 4.7 on September 21 across Grok Build, Cursor, the API, and OpenRouter, and the number that matters isn't the benchmark score — it's the invoice. Grok 4.7 lands at Grok 4.6 pricing while ranking second only to Anthropic's own models on Artificial Analysis's AA-Briefcase, with its Elo jumping from 1698 to 1994 and matching Opus 5 on certain tasks at roughly half the per-task cost.
That's a direct shot at the price story we've been tracking for weeks: Anthropic has held Sonnet 5 at its introductory $2/$10 rate through repeated pressure, treating price stability as a point of pride even while quietly cutting Claude Code capacity 17% through the expired usage boost. The Sonnet freeze made sense when nothing else at that quality tier was cheaper. Grok 4.7 breaks that logic — it's not undercutting Anthropic on a lesser model, it's undercutting Opus-class output specifically, and it's doing it on OpenRouter, the same neutral layer where solo builders already compare every option side by side.
For anyone routing a coding agent or analytical workload through Opus 5 right now, this is a live, callable alternative today, not a future promise: same interface surfaces (Grok Build, Cursor, API, OpenRouter), roughly half the cost per task, and a published benchmark position ahead of everything except Anthropic. It doesn't force Anthropic to cut Sonnet's price this week, but it does raise the cost of holding the freeze — every week Grok 4.7 sits at this price-to-capability ratio is a week Anthropic's stability pitch looks more like inertia and less like discipline. Solo builders don't need to wait for that decision to play out; the swap is available now, and the savings are real enough to test against your own workload rather than take on benchmark faith alone.
More this week:
|
