SpaceXAI Launches Grok 4.7 for Coding and Knowledge Work

SpaceXAI has launched Grok 4.7, describing it as its most capable model for coding and knowledge work. It works longer on difficult tasks and checks its own work more carefully.

The model is available today in Cursor and Grok Build, as well as through the Grok API, third-party coding harnesses, and model routers and cloud platforms, according to SpaceXAI. Pricing starts at $2 per million input tokens and $6 per million output tokens, matching Grok 4.6 in the comparison table while listing lower input and output rates than GPT-5.6 Sol and Fable 5.1, with a fast variant offering twice the output speed at twice the price.

It uses a new, larger base model than its predecessor, trained with a longer reinforcement learning run on a harder mix weighted toward multi-hour problems. The model is better at verifying its own work and managing longer context, plus native understanding of the Grok Bot harness for conversational tasks and general knowledge work. For documents and presentations, the announcement points to gains on GDPval and AA Briefcase, with the latter rising to 1,657 from 1,546, just short of the 1,678 listed for Fable 5.1.

On CursorBench 4.0, which stresses longer-running coding tasks, Grok 4.7 scores 46.3% in its table, up from 40.4% for Grok 4.6 and ahead of the 41.7% listed for GPT-5.6 Sol, though behind the 51.8% listed for Fable 5.1. It also posts 64.0% on EEBench and 19.6% on the Harvey legal benchmark, both ahead of the other three models in its table, while Terminal-Bench remains higher for Fable 5.1 at 57.9% against 38.0% for Grok 4.7, up from 20.3% for Grok 4.6.

Grok 4.7 is described as the strongest model the company has tested on refusals and jailbreak resistance, built with an entirely new safeguard stack. In its announcement, it tops LatchBio's biosafety benchmark at 62.4% and allows only 3.3% of risky dual-use prompts through on HackerBench v0.3, while retaining low refusal rates for legitimate security work. Select cybersecurity partners are getting invite-only access to red-team capabilities for defense research.