Grok 4.7 puts xAI in the top four - and it buys that with three times the tokens

xAI shipped Grok 4.7 today. Our radar read the id at 15:49:44 UTC, confirmed it on the next sweep and sent 307 alerts 61 seconds after first sight.

The spec sheet barely moves. From xAI's own docs: grok-4.7, 500,000-token context, $2 in / $6 out per million under a 200k prompt and $4 / $12 over it, cached input $0.50, text and image in, reasoning effort low through xhigh. Every one of those numbers is identical to Grok 4.6. The gateways are discounting - OpenRouter lists it at $1.60/$4.80 - but the list price did not move.

What moved is the ranking. Artificial Analysis puts Grok 4.7 at 46 on its Intelligence Index, up 2 points, which lifts xAI into the top four labs. On long-horizon agentic knowledge work the jump is bigger: 1657 Elo on AA-Briefcase, up 111 over Grok 4.6, which places it alongside Claude Opus 5 and Claude Fable 5.1 at the frontier, and 1695 on GDPval-AA, up 90. With Grok Build it scores 56 on the Coding Agent Index, up 9, ranking 4th among models in their native harnesses - behind only Fable 5.1, GPT-6 Astra and Opus 5.

Outside agentic work it is close to flat, and not uniformly up: Terminal-Bench 4.0 improves 4.5 points and GDP.pdf 3.0, while AA-LCR drops 3.7 and AutomationBench-AA 1.1.

The cost of those gains is tokens. At xhigh, Artificial Analysis measured about 81,000 output tokens per Intelligence Index task, against 36,000 for Grok 4.6 at high and 27,000 for GPT-6 Astra at max - 125% and 196% more. Run that against the list prices and the picture is less obvious than "it got expensive": at $6 per million output, 81k tokens is about $0.49 of output per task. The same task on Grok 4.6 was about $0.22, so 4.7 costs roughly 2.25x its predecessor. But GPT-6 Astra, at a third of the tokens and $50 per million, is about $1.35 - nearly three times Grok 4.7. Grok is burning far more tokens to get there and is still the cheaper way to get there.

That is the trade to check against your own workload: if you are paying per token, 4.7 more than doubles the bill against 4.6 for a 2-point index gain, and the gain is concentrated in agentic work. If you are paying per task and comparing against the frontier, it is the cheapest seat at that table.

On our end: this hunt had been dead for five weeks. Its one-shot had been spent on Grok 4.6 back in August, so the entry could not fire; we moved that event to Grok 4.6, where it belonged, this morning. The launch landed hours later and the alert went out automatically.