Musk reposted a tweet from xAI engineer Eric Zakariasson, unveiling the model card for Grok 4.7. The data shows that Grok 4.7 improved from 20.3% to 38.0% on the Terminal-Bench benchmark, from 31.9% to 46.0% on SWE-Marathon, from 48.5% to 56.7% on HealthBench Pro, from 15.8% to 19.6% on Legal Agent, and from 60.0% to 66.0% on EEBench. The new Grok 4.7 version will be offered at the same price as Grok 4.6.