Extended task handling
A longer reinforcement-learning run focused on harder, multi-hour problems is intended to improve performance on tasks that require sustained work rather than a short answer.
Grok 4.7
Freemium
Grok 4.7 is an AI model for software development and knowledge work, designed for extended coding and professional tasks. It is available through Grok Build, Cursor, the Grok API, and other supported coding and model-serving platforms.
什么是 Grok 4.7？
Grok 4.7 is an AI model for software engineering and knowledge work. It is designed for difficult, longer-running tasks, with improvements in extended-context management, self-checking, and work that may continue for several hours. The model is also trained to understand the Grok Bot harness for conversational tasks and general knowledge work.
Its published evaluation coverage includes coding, terminal work, electrical engineering, office and professional knowledge work, legal work, clinical reasoning, and safety-sensitive cybersecurity and biological tasks. Grok 4.7 is available through Grok Build, Cursor, the Grok API, third-party coding harnesses, model routers, and cloud platforms. The launch page lists usage starting at $2 per million input tokens and $6 per million output tokens.
Grok 4.7 能做什么？
Work verification and context management
Grok 4.7 is described as better at checking its own work and managing longer context, which is relevant to extended coding, terminal, and document workflows.
Coding and professional work evaluation
The published results cover software engineering, terminal work, electrical engineering, office work, legal work, and clinical reasoning, including a 46.3% CursorBench 4.0 score and a 1,657 AA Briefcase v1.1 score.
Document and presentation creation
The launch material reports improved performance in creating documents and presentations, with comparisons on GDPval and AA Briefcase against earlier and other frontier models.
Safety and cybersecurity safeguards
A new safeguard stack is intended to improve refusal behavior and jailbreak resistance while supporting legitimate cybersecurity work. Select cybersecurity partners can access red-team capabilities by invitation.
使用场景
“Long-running software projects”
Developers can use Grok 4.7 for coding tasks that require sustained reasoning, including work evaluated by CursorBench and DeepSWE, rather than limiting it to short code-generation requests.
“Terminal and engineering workflows”
Teams working with terminal-based tasks or electrical-engineering problems can evaluate the model against their own workflows, reflecting the model’s published Terminal-Bench and EEBench coverage.
“Professional documents and presentations”
Knowledge workers can use the model to help create business documents and presentations, a workflow represented by the reported GDPval and AA Briefcase results.
“Security research and defensive cyber work”
Security practitioners can apply the model to legitimate cybersecurity and defense research. Access to the described red-team capabilities is limited to select partners by invitation.
常见问题
Where can I use Grok 4.7?
The launch page says Grok 4.7 is available in Cursor and Grok Build, through the Grok API, and through third-party coding harnesses, model routers, and cloud platforms.
How is Grok 4.7 priced?
The listed starting rates are $2 per million input tokens and $6 per million output tokens. A fast variant provides twice the output speed at twice the price. The source does not specify broader plan limits or usage quotas.
What types of work is Grok 4.7 intended for?
Its stated focus is coding and knowledge work, with published evaluations covering software engineering, terminal work, electrical engineering, professional office work, legal work, clinical reasoning, and safety-sensitive cybersecurity and biological tasks.
Does Grok 4.7 support cybersecurity work?
Yes. The source describes utility for legitimate cybersecurity work alongside safeguards for risky or malicious tasks. Select cybersecurity partners receive invite-only access to red-team capabilities for defense research.
快速信息
- Product type
- AI model for coding and knowledge work
- Available through
- Grok Build, Cursor, Grok API, coding harnesses, model routers, and cloud platforms
- Starting token price
- $2 per million input tokens; $6 per million output tokens
- Fast variant
- Twice the output speed at twice the listed price
- Published coding benchmark
- 46.3% on CursorBench 4.0
- Launch date
- September 21, 2026
Grok 4.7 替代品
Amazon Q Developer
aws.amazon.com
Amazon Q Developer is an AI-powered coding assistant for software developers. It provides code suggestions, chat, testing, security scanning, refactoring, and agentic help across supported IDEs, the command line, AWS tools, and selected collaboration workflows.
Verdent AI
verdent.ai
Verdent AI 是一套智能代理编程工具，可将自然语言目标转化为任务、构建和测试结果，支持 VS Code、桌面应用、Mac、Windows、Slack 和 Telegram。
AutonomyAI
autonomyai.io
AutonomyAI is an AI-native product delivery platform that helps product managers and designers turn ideas into production-ready code changes in their existing codebase. Engineers review and approve the resulting pull requests before merge.
Gemini 3.7 Flash
deepmind.google
Gemini 3.7 Flash is a Google AI model for software engineering, web development, knowledge work, and agent workflows. It is available to developers through Google AI Studio, the Gemini API, Android Studio, and Google Antigravity, with enterprise and Gemini app access also described by Google.
Freebuff
freebuff.com
Freebuff is an ad-supported coding-agent suite for building and modifying software through the CLI, desktop app, web builder, or cloud workspace. It also includes an AI chat product for research and in-depth answers, with free access and optional paid plans.
ZZZ Code AI
zzzcode.ai
在线使用 AI 工具生成、审查、转换和解释多种编程语言代码