xAI Releases Grok 4.7… Enhances Coding and Knowledge Work Performance

xAI has released Grok 4.7, positioning it as the strongest model for coding and knowledge work. It uses a larger base model compared to Grok 4.6 and underwent long RL training focused on difficult tasks taking several hours to improve performance.

Performance and Pricing

Grok 4.7 features improved self-task verification and long-context management capabilities. In particular, it natively understands the Grok Bot harness, enhancing conversational and general knowledge work abilities. Pricing is identical to Grok 4.6 and is competitive compared to models in the same tier.

• Token pricing: Input $2, Output $6 (per million tokens)
• Benchmarks: Recorded 46.3% on CursorBench 4.0, achieving price-performance frontier in long coding tasks.
• Other metrics: Significant improvements over the previous version were confirmed on Terminal-Bench 4.0 (38.0%), Harvey Legal Agent Benchmark (19.6%), etc.

Safety and Availability

A completely new safeguard stack was applied to secure the strongest safety in refusal and jailbreak resistance tests. It took the lead in dual-use fields (cybersecurity, biology) in both the utility of safe tasks and the safe refusal of dangerous tasks.

• Safety metrics: Ranked 1st on the LatchBio biosafety benchmark with 62.4%, and recorded the highest safety with a 3.3% acceptance rate for dangerous prompts on HackerBench v0.3
• Availability: Immediately available on Cursor, Grok Build, Grok API, third-party coding harnesses, model routers, and cloud platforms
• Additional options: A Fast variant is offered with 2x faster output speed and 2x the price