xAI has released Grok 4.7, its latest flagship artificial intelligence model developed for multi-hour coding, technical reasoning, and professional knowledge workflows. The update focuses on extended task execution, self-verification of generated code, and an overhauled safety architecture. According to the official xAI announcement, the model maintains identical base pricing to its predecessor while achieving higher benchmark scores across multiple domains.
Rather than solely scaling model size, xAI structured the training run around problems that require sustained reasoning over several hours. This adjustment improves context retention, multi-file code editing, and autonomous error correction during agentic tasks.
Grok 4.7: xAI foundation model for software engineering, professional knowledge work, and dual-use safety.
Architecture and training improvements
Grok 4.7 builds on a larger base architecture than Grok 4.6. The engineering team applied a longer reinforcement learning schedule trained against complex task distributions. A substantial portion of the training dataset consisted of problems that take human specialists many hours to solve, such as large codebase refactoring, electrical circuit design, and legal statutory analysis.
A central upgrade in Grok 4.7 is internal verification. During complex multi-step tasks, the model evaluates its intermediate code logic, inspects terminal outputs, and checks for syntax errors before presenting a final answer. The system also integrates natively with the Grok Bot harness, improving dialogue coherence and tool coordination across conversational workflows.
Software engineering and terminal benchmarks
In software development, Grok 4.7 demonstrates notable price-to-performance efficiency. On CursorBench 4.0, a benchmark measuring long-horizon coding tasks within the Cursor code editor, Grok 4.7 achieves parity with leading proprietary frontier models at roughly half the execution cost.
CursorBench 4.0 performance versus cost per task, comparing Grok 4.7 with Opus 5, Fable 5.1, GPT-5.6 Sol, and Sonnet 5.
At the Extra High effort setting, Grok 4.7 scores 46.3% accuracy on CursorBench 4.0 with an average task cost of $6.01. For comparison, Opus 5 reaches 46.6% accuracy but incurs an average cost of $11.95 per task. Grok 4.7 also surpasses GPT-5.6 Sol Max (41.7% at $8.23) and Sonnet 5 Max (34.1% at $7.17).
Model & Configuration
CursorBench 4.0 Score
Average Cost per Task
Grok 4.7 (Extra High)
46.3%
$6.01
Grok 4.7 (High)
43.9%
$4.69
Grok 4.7 (Medium)
41.6%
$3.49
Grok 4.7 (Low)
33.1%
$1.58
Opus 5 (Max)
46.6%
$11.95
Opus 5 (High)
44.7%
$9.00
GPT-5.6 Sol (Max)
41.7%
$8.23
Fable 5.1 (Max)
51.8%
$17.28
Sonnet 5 (Max)
34.1%
$7.17
Accuracy and cost comparison on CursorBench 4.0 software engineering tasks.
Beyond editor completions, Grok 4.7 records solid results in autonomous agent benchmarks. On DeepSWE v1.1, which tests multi-file code fixes, Grok 4.7 reaches 71.0% in high effort mode, comparable to GPT-5.6 Sol Max (72.7%) and Fable 5.1 Max (70.0%). On Terminal-Bench 4.0, which evaluates multi-hour command-line interaction and environment debugging, Grok 4.7 achieves 38.0%, almost doubling the 20.3% score of Grok 4.6.
Professional knowledge work and domain expertise
xAI evaluated Grok 4.7 across several non-coding professional domains, including electrical engineering, office workflows, legal analysis, and clinical medicine:
Electrical engineering (EEBench): Grok 4.7 achieves 64.0%, leading all evaluated models in circuit design and hardware problem solving, ahead of Fable 5.1 (56.4%) and Grok 4.6 (53.0%).
Document and presentation analysis (GDPval): Grok 4.7 scores 1,695 Elo points, placing second behind Fable 5.1 (1,735 Elo) and exceeding GPT-6 Astra (1,542 Elo).
Multi-hour office work (AA Briefcase v1.1): The model earns 1,657 Elo points, improving over Grok 4.6 (1,546 Elo) and GPT-5.6 Sol (1,487 Elo).
Legal analysis (Harvey Legal Agent Benchmark): Grok 4.7 leads the benchmark at 19.6%, outperforming Grok 4.6 (15.8%), Fable 5.1 (6.7%), and GPT-5.6 Sol (2.5%).
Clinical reasoning (HealthBench Professional): Grok 4.7 records 56.7%, up from 48.5% in Grok 4.6.
Elo scores on the GDPval professional knowledge benchmark across frontier AI models.
Benchmark / Metric
Grok 4.7 (xHigh)
Grok 4.6 (High)
GPT-5.6 Sol (Max)
Fable 5.1 (Max)
Input Token Price ($/M)
$2
$2
$4
$10
Output Token Price ($/M)
$6
$6
$20
$50
CursorBench 4.0
46.3%
40.4%
41.7%
51.8%
DeepSWE v1.1
71.0%*
65.2%
72.7%
70.0%
EEBench (Electrical Eng.)
64.0%
53.0%
39.4%
56.4%
AA Briefcase v1.1
1,657
1,546
1,487
1,678
Terminal-Bench 4.0
38.0%
20.3%
37.3%
57.9%
Harvey Legal Agent
19.6%
15.8%
2.5%
6.7%
HealthBench Professional
56.7%
48.5%
60.5%
62.1%
Comprehensive benchmark results and API pricing for Grok 4.7 compared to frontier AI systems.
Safety, cybersecurity, and dual-use protection
Grok 4.7 incorporates an updated safety stack designed to separate hazardous requests from legitimate technical work. In dual-use evaluations, the model demonstrated improved calibration between safety refusals and domain utility.
On the LatchBio biosafety benchmark, Grok 4.7 achieved 62.4%, the highest recorded score for blocking dangerous biological synthesis protocols while assisting with standard bioengineering workflows. In cybersecurity evaluations on HackerBench v0.3, the model allowed only 3.3% of risky malicious prompts through while maintaining low refusal rates for defensive vulnerability research. xAI has also begun granting select defensive security teams access to Grok 4.7 red-team tools for infrastructure protection.
Pricing, latency, and deployment options
Grok 4.7 is accessible through multiple platforms. Developers can test the model through the Grok web interface, build autonomous applications inside Grok Build, or access the REST API directly. Base pricing remains fixed at $2.00 per million input tokens and $6.00 per million output tokens. A dedicated fast variant is also available, providing double the output generation speed at $4.00 per million input tokens and $12.00 per million output tokens.
