Introducing Grok 4.7

What changed

Source summary

xAI released Grok 4.7, its latest flagship model for coding and knowledge-work tasks, built on a larger base model trained on more challenging data. It improves long-context handling and self-verification and adds an enhanced safety/security safeguard framework, while keeping pricing and speed in line with Grok 4.6.

Why it matters to your work

Practical interpretation

If your team uses Grok for coding or research tasks, Grok 4.7 offers stronger reasoning and long-context handling at the same price as the previous version, worth trying for complex multi-step work like codebase reviews or in-depth report drafting. Most relevant to technical staff already using xAI's tools rather than casual chat users.

When this could help

Worth evaluating if you currently use a more expensive frontier model for coding, document creation, or multi-hour office work and want to test whether Grok 4.7 delivers comparable quality at lower cost.

Grok 4.7 is xAI's most capable model for coding and knowledge work at $2/$6 per 1M tokens (same as Grok 4.6), matching or exceeding GPT-5.6 Sol and Fable 5.1 on several benchmarks at a fraction of their price. It is trained to work longer on hard tasks, verify its own output, and natively understand the Grok Bot harness.

How to get started

1. Try Grok 4.7 for free at x.ai/build (Grok Build), or select it in Cursor or your preferred coding harness.

   Provider guidance ↗

2. For API usage, use the Grok API. For speed-sensitive tasks, select the Fast variant at 2x price for 2x output speed.

   Provider guidance ↗

Build it into your routine

SCALA suggested workflow

1. Coding model comparison

   Run a representative coding or document-creation task through Grok 4.7, your current model, and one cheaper alternative. Compare quality and cost per task to decide whether switching saves money without sacrificing quality.

What to check

• The announcement does not list rate limits, context window size, or usage caps. Source ↗
• Red-team cybersecurity access is invite-only for select partners. Source ↗
• Benchmark results are vendor-reported; some competitor scores use different effort levels or may not have Fable 5.1 equivalents. Source ↗