xAI Launches “grok-code-fast-1” for Agentic Coding
Elon Musk’s AI startup xAI unveiled grok-code-fast-1, its first dedicated agentic coding model designed from scratch for speed and cost-efficiency in real-world developer workflows.
xAI describes grok-code-fast-1 as a purpose-built code assistant optimized for IDE integrations, automated coding agents, and large-repo reasoning.
Unlike generalist language models, this variant focuses on delivering fast, economical responses with minimal developer oversight while supporting complex, multi-step coding workflows.
Key Features
Massive Context Window: Enables the model to ingest entire repositories or multi-file projects in a single session, facilitating coherent interactions across extensive codebases.
Agentic Coding & Tool Use: Supports function calling and structured outputs (e.g., JSON diffs), allowing it to act as an “agent” that invokes tools, chains operations, and returns machine-friendly responses for CI/CD pipelines and automation scripts.
Speed and Throughput: Processes up to 160–190 tokens per second, outperforming competitors like Google Gemini 2.5 Pro (92 TPS) and Qwen3-Coder (80 TPS).
This high throughput makes interactive coding loops feel instantaneous.
Visible Reasoning Traces: Surfaces intermediate “thinking” steps to enhance steerability and debugging transparency, helping developers inspect and guide the model’s decision process.
Developer-Oriented Outputs: Optimized for code generation, debugging, and stepwise reasoning, the model excels at boilerplate creation, bug fixes, and refactoring tasks with robust accuracy in coding-specific benchmarks.
Broad Language Support: Proficient in TypeScript, Python, Java, Rust, C++, and Go, covering the full software stack and major enterprise languages.
Benchmark Performance
In internal tests, grok-code-fast-1 scored 70.8% on the SWE-Bench Verified benchmark and demonstrated strong reliability in coding accuracy, morality, email classification, and general knowledge tasks.
Availability & Pricing
Limited-Time Free Access: For launch, xAI is offering free access through select partners, including GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf.
GitHub Copilot Public Preview: Rolling out as an opt-in preview for Copilot Pro, Pro+, Business, and Enterprise users via Visual Studio Code.
Until September 2, 2025, it’s available complimentary; afterward, access continues via normal Copilot pricing or the Bring-Your-Own-Key (BYOK) route with xAI API key.
Direct API Access & Pricing: After the free period, pricing is competitive at:
- $0.20 per million input tokens.
- $1.50 per million output tokens.
- $0.02 per million cached input tokens.
- xAI also offers generous rate limits (~480 requests/minute, ~2 million tokens/minute).
What’s Next
xAI is committed to rapid iteration, with updates expected in days, not weeks.
A next-gen variant featuring multimodal inputs, parallel tool-calling, and extended context support is already in training.
News Gist
xAI has launched grok-code-fast-1, a specialized AI model for agentic coding, designed to accelerate code generation, debugging, and automation.
Optimized for speed, scalability, and developer efficiency, it marks xAI’s bold push into AI-driven programming tools.
FAQs
Q1: What is grok-code-fast-1?
A1: xAI’s high-speed, low-cost coding AI model optimized for agentic workflows.
Q2: When was it announced?
A2: August 28, 2025.
Q3: What makes it fast?
A3: Processes 160–190 tokens per second with optimized inference.
Q4: How large is its context window?
A4: 256 000 tokens—enough to load entire codebases.
Q5: How can developers access it?
A5: Free preview via GitHub Copilot; API on xAI and partners.
Q6: What is the pricing?
A6: $0.20 per 1M input tokens; $1.50 per 1M output tokens.
