xAI launches agentic coding model that is fast and cheap
The artificial intelligence company xAI has announced the launch of a new AI model — grok-code-fast-1 — signaling its entry into the agentic coding market. The company touts the model as “speedy daily driver” for developers who find other powerful models “frustratingly slow” for day-to-day tasks. xAI says the model is over 60% (160 TPS) faster than the fastest competitors available: Gemini 2.5 Pro (92.39 TPS) and Qwen3-Coder (80 TPS).
The company is also positioning it to be an economical choice, pricing it at $0.20 per million input tokens and $1.50 per million output tokens. Per the announcement, the model is versatile across the full software stack and is especially proficient at languages like TypeScript, Python, Java, and Rust.
In terms of performance, xAI reports that grok-code-fast-1 scored 70.8% on SWE-Bench-Verified benchmark in its internal tests. However, the company says its focus is on real-world usability, guided by human assessments from experienced developers who have praised the model for being “fast and reliable” for day-to-day coding.
The model, which xAI initially released secretly under the codename “sonic,” is now generally available via API. You can also use the model for free for a limited time through xAI's launch partners, including GitHub Copilot, Cline, and Cursor. xAI also stated that a new, more advanced variant that supports multimodal inputs, extended context length, and parallel tool calling is already in the works.
