xAI Launches Grok 4.7 With a New Safeguard Stack, Same Pricing as Grok 4.6

xAI released Grok 4.7, built primarily for coding and professional knowledge work, with a larger base model trained through extended reinforcement learning on more demanding tasks.

Details

• Core behavior: works longer on difficult tasks and checks its own work more carefully than Grok 4.6, with improved context management and self-verification, and native integration with the Grok Bot conversational interface
• Benchmarks: scores 46.3% on CursorBench 4.0 (coding), 19.6% on the Harvey legal benchmark, 64.0% on an electrical engineering benchmark, and 56.7% on clinical reasoning — with meaningful gains over Grok 4.6 in document creation, professional office work, and terminal operations at similar speed
• Safety changes: xAI describes an “entirely new safeguard stack” with enhanced jailbreak resistance and balanced cyber capabilities, scoring 62.4% on biosafety benchmarks
• Pricing: $2 per million input tokens and $6 per million output tokens, matching Grok 4.6’s pricing exactly
• Availability: accessible through Cursor, Grok Build, the Grok API, and third-party platforms, with a free trial at x.ai/build

What happened next

Grok 4.7 arrives with no price increase over its predecessor despite the added capability, and lands the same week as GPT-6 Sol and Luna and Claude Opus 5.5 — all three major AI labs shipping mid-cycle model updates within days of each other.