SpaceXAI said Grok 4.7 is now available for coding and knowledge work through Cursor, Grok Build, the Grok API and third-party platforms.. The company said the model starts at $2 per million input tokens and $6 per million output tokens, with a faster version offering twice the output speed at twice the price..
SpaceXAI described Grok 4.7 as its most capable model for coding and knowledge work, and said it runs at the same price and speed as Grok 4.6 while using a larger base model.. The company said Grok 4.7 was trained with a longer reinforcement learning run on a harder mix of tasks, with greater weighting toward problems that take many hours to complete..
SpaceXAI said the model is better at verifying its own work, handling longer context and understanding the Grok Bot harness, which it said improves conversational tasks and general knowledge work.. On CursorBench 4.0, which tests longer-running coding tasks, SpaceXAI said Grok 4.7 was at the frontier in price-performance.
The company said Grok 4.7 also improved on Grok 4.6 in document and presentation creation, and posted gains on the GDPval and AA Briefcase benchmarks, which test work associated with professions including law, nursing and financial analysis.. SpaceXAI said Grok 4.7 was built with a new safeguard stack and was the strongest model it had tested for refusals and jailbreak resistance..
In dual-use fields including cybersecurity and biology, the company said the model balanced utility for benign tasks with refusals on dangerous ones, and scored 62.4% on LatchBio’s biosafety benchmark.. On HackerBench v0.3, SpaceXAI said Grok 4.7 allowed 3.3% of risky dual-use prompts through while rarely blocking legitimate security work.
The company said it has started giving select cybersecurity partners invite-only access to Grok 4.7 red-team capabilities for defense research.. Featured image credit. Tags: grokSpaceXAI
