Grok 4.7 Release: xAI Redefines the Frontier of Autonomous Coding Agents

In specialized benchmarks, Grok 4.7 demonstrates professional-grade proficiency. It achieved 46.3% on CursorBench 4.0, a massive improvement over Grok 4.6's 40.4%, placing it at the absolute frontier of price-performance for long-running tasks. Its ability to handle high-effort engineering is further validated by a 71.0% score on DeepSWE v1.1. Beyond code, it shows remarkable versatility with scores like 64.0% on EEBench and 19.6% on the Harvey Legal Agent Benchmark, suggesting a level of reasoning comparable to specialized professionals.

• CursorBench 4.0: 46.3%
• DeepSWE v1.1 (High Effort): 71.0%
• Coding Agent Index: 56 (up from 47)
• Terminal-Bench 4.0: 38.0%
• GDPval: 1695 Elo

Safety and Robustness: The New Safeguard Stack

As models become more capable of executing code and navigating systems, safety becomes paramount. xAI has deployed an entirely new safeguard stack with Grok 4.7, making it the strongest model tested to date regarding refusal accuracy and jailbreak resistance. This is critical for enterprise deployment where security is non-negotiable.

The model's performance on HackerBench v0.3 is particularly noteworthy: it allows only 3.3% of risky dual-use prompts through, yet it is specifically tuned to avoid the 'false refusal' trap that plagues many competitors, ensuring it rarely blocks legitimate security research or defensive coding work. Additionally, it tops LatchBio's biosafety benchmark with a score of 62.4%, demonstrating a sophisticated understanding of high-stakes domain boundaries.

• HackerBench v0.3: Only 3.3% risky prompt leakage
• LatchBio Biosafety Benchmark: 62.4%
• Optimized to minimize false refusals in legitimate security workflows

API Pricing and Availability

xAI has maintained a highly competitive pricing strategy, offering Grok 4.7 at the same base price as Grok 4.6. This provides developers with a massive intelligence upgrade at zero additional cost for standard workloads. For users requiring extreme throughput, a fast variant is available, offering twice the output speed at twice the price.

The model is widely accessible through various channels, including direct integration in Cursor and Grok Build, as well as the official Grok API. It is also compatible with third-party coding harnesses, model routers, and major cloud platforms, making it easy to integrate into existing CI/CD pipelines and developer environments.

• Standard Input: $2.00 / MTok
• Standard Output: $6.00 / MTok
• Cache Hit Input: $0.50 / MTok
• Fast Variant: 2x output speed at 2x cost

Use Cases: Where to Deploy Grok 4.7

Grok 4.7 is best utilized in scenarios requiring deep reasoning and long-term planning. For software engineering teams, it is an ideal candidate for autonomous bug fixing, large-scale refactoring, and complex architectural migrations where the model must 'verify its own work' through its specialized RL training.

Beyond pure coding, its high performance in legal (Harvey Benchmark) and medical (HealthBench Professional: 56.7%) domains makes it a powerful tool for professional knowledge work, RAG (Retrieval-Augmented Generation) systems involving massive technical documentation, and sophisticated agentic workflows that require navigating terminal environments and complex file structures.

• Autonomous Software Engineering (Agentic workflows)
• Large-scale codebase refactoring and maintenance
• Professional Knowledge Work (Legal, Medical, Technical)
• Advanced RAG with 500K context window

Getting Started

To begin using Grok 4.7, developers can access it immediately via the Grok API or through integrated IDE extensions like Cursor. If you are building custom agentic tools, the Grok API provides the necessary flexibility to leverage the high-effort reasoning modes and the massive context window.

For those looking for a managed experience, Grok Build offers a streamlined environment to prototype and deploy applications powered by Grok 4.7's advanced reasoning capabilities.

• Direct API access via xAI developer portal
• Integrated support in Cursor and Grok Build
• Compatible with third-party model routers and cloud providers

API Pricing — Input: $2.00 per million tokens / Output: $6.00 per million tokens / Context: 500K tokens

Sources