SpaceXAI launches Grok 4.7 with a focus on coding and complex professional tasks

SpaceXAI has released Grok 4.7, its new flagship AI model for coding and professional knowledge work, with a larger base model, longer reinforcement learning runs and the same API pricing as Grok 4.6.

The company says Grok 4.7 was trained on a harder mix of tasks, with more emphasis on problems that can take hours to complete. Compared with Grok 4.6, the model is designed to verify its own work more carefully, handle longer context more effectively and natively understand the Grok Bot harness. On GDPval, which evaluates work such as presentations and documents produced for professional tasks, Grok 4.7 scored 1,695, compared with 1,542 for GPT-6 Astra.

Coding performance has also improved. Grok 4.7 scored 46.3% on CursorBench 4.0, ahead of GPT-5.6 Sol at 41.7%. On DeepSWE v1.1, the model reached 71% at high reasoning effort.

SpaceXAI also introduced an entirely new safeguard stack for Grok 4.7, which the company says improves resistance to jailbreaks while preserving access to legitimate research tools. The model scored 62.4% on LatchBio's biosafety benchmark. On HackerBench v0.3, a benchmark covering risky and malicious cybersecurity tasks, it allowed 3.3% of risky dual-use prompts through. SpaceXAI has also given selected cybersecurity partners invite-only access to Grok 4.7's red-team capabilities for defensive research.

API pricing remains unchanged from Grok 4.6 at $2 per million input tokens and $6 per million output tokens. By comparison, SpaceXAI lists GPT-5.6 Sol at $4 and $20 per million input and output tokens, respectively. A fast version of Grok 4.7 offers twice the output speed for twice the price.

Grok 4.7 is available now in Cursor and Grok Build, as well as through the Grok API, third-party coding tools, model routers and cloud platforms.