xAI announced and released Grok 4.7, an upgraded model built on a larger base and extended reinforcement learning aimed at complex, long-horizon coding and knowledge work. xAI reports meaningful improvements over Grok 4.6 across multiple benchmarks: CursorBench 4.0 scored 46.3% versus 40.4% for 4.6, and Terminal Bench 4.0 rose to 38% from 20.3%. The company also cited 71% on DeepSWE 1.1 at high reasoning effort, 1,657 on AA Briefcase and 64% on EEBench, and noted that Grok 4.7 leads GPT 5.6 Sol on several evaluations while other models remain ahead on different tests.
The model runs at 2.1 trillion parameters, about 40% more than Grok 4.6’s 1.5 trillion, and incorporates supplemental training data from SpaceX engineering sources including Starlink telemetry, manufacturing records and failure logs. Pricing is unchanged from Grok 4.6: $2 per million input tokens and $6 per million output tokens, with a faster output tier at twice the price. Grok 4.7 is available immediately via the Grok app, Cursor, Grok Build and the xAI API, plus third-party coding tools and cloud platforms.
xAI said the model spends longer on hard problems and self-verifies more often, and introduced an enhanced safeguard stack; HackerBench 0.3 allowed 3.3% risky dual-use prompts through, and a LatchBio biosafety benchmark scored 62.4%. The release followed multiple delays; Elon Musk described Grok 4.7 as a notable improvement that combines intelligence, speed and low cost, while competitors like Claude Fable 5.1 and GPT-6 Astra scored higher on some frontier benchmarks. xAI is offering red-team access to selected cybersecurity partners for defensive research.
This summary is composed by the cFlash AI agent from multiple public sources, under human supervision. The content is for informational purposes only and does not constitute investment, financial, legal, or tax advice.