News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
fnGrok Code Fast 1

Grok Code Fast 1

A fast, low-cost reasoning model built for agentic coding work.

Grok Code Fast 1 is a reasoning model aimed at agentic coding, where an AI plans, calls tools and edits code in loops. xAI built it from scratch and released it in 2026 after a stealth phase under the codename sonic, pricing it cheaply and offering it free through partner coding tools.

Share on WhatsApp

  • Built from scratch with a new architecture for agentic coding
  • Scored 70.8% on SWE-Bench-Verified in xAI's own harness
  • Launched quietly under the codename sonic before public release
  • Priced at $1.50 per million output tokens and $0.02 per cached input
  • xAI reports prompt cache hit rates above 90% with launch partners

What it is

Grok Code Fast 1 is a reasoning model that xAI made for agentic coding: the style of programming where an assistant works through a task by thinking, running commands and editing files on its own. The company says it built the model from scratch, with a new architecture and training data weighted toward programming. It handles everyday developer work across the stack, from starting a project to answering questions about a codebase and making precise bug fixes. xAI highlights TypeScript, Python, Java, Rust, C++ and Go. It is priced low and offered through an API and partner coding tools.

How it came about

xAI's engineers, who use agentic coding tools heavily, wanted something quicker and cheaper for their own daily work, so they started a new model rather than adapting an existing one. The model first appeared quietly under the codename sonic, and the team watched community channels and shipped several checkpoints in response to feedback. It was then announced publicly as grok-code-fast-1. xAI says updates will come in days rather than weeks, and that a variant handling multimodal input, parallel tool calls and a longer context window is already being trained.

Inside the model

The model was trained in two stages. Pre-training used a corpus deliberately filled with programming material; post-training used curated data modelled on real pull requests and coding tasks. xAI worked with launch partners to tune how the model behaves inside their agentic platforms, and it learned common developer tools such as grep, a terminal and file editing. The company's inference and supercomputing teams added serving techniques to speed up responses, plus prompt caching work that it says regularly reaches cache hit rates above 90% with partner tools.

What it can do

xAI reports a score of 70.8% on the full SWE-Bench-Verified subset, measured with its own internal harness. The company cautions that benchmarks like SWE-Bench miss the feel of real software work, so it also runs routine human reviews, with experienced developers rating end-to-end performance, and automated checks on behaviour. Programmers, it says, judged the model fast and dependable for routine coding. A user quoted by xAI built a battle simulator prototype in under a day, praising the response speed and instruction following, and said small focused tasks work better than one large prompt.

How to use it

The model is served through the xAI API at $0.20 per million input tokens, $1.50 per million output tokens and $0.02 per million cached input tokens. For a limited period it was also free in partner tools: GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode and Windsurf. GitHub Copilot described early testing as showing both speed and quality on agentic coding work.

Reception and debate

Partner quotes published by xAI praise the model's speed and its quality on agentic coding tasks. A developer using it in Cursor called it the fastest model available and said its speed changed how they work, preferring to hand it small tasks and iterate quickly. xAI's own framing is that programmers rate it fast and reliable for everyday work.

Why it matters

xAI presents the model as the start of a new family and says it will iterate rapidly on developer feedback, with improvements arriving within days. The company also notes a follow-up variant in training that adds multimodal inputs, parallel tool calling and a longer context window.

Questions people ask about Grok Code Fast 1

What is Grok Code Fast 1 for?

It is a reasoning model aimed at agentic coding, where an assistant plans, calls tools and edits files in loops. xAI says it handles everyday developer tasks across the stack, from starting new projects to answering codebase questions and making precise bug fixes, with particular strength in TypeScript, Python, Java, Rust, C++ and Go.

Who made it and when did it appear?

xAI built it from scratch. It first went out quietly under the codename sonic, with the team watching community channels and shipping several checkpoints, and was then announced publicly as grok-code-fast-1. xAI says a multimodal variant with parallel tool calls and a longer context window is already in training.

How do I get it and what does it cost?

It is available through the xAI API at $0.20 per million input tokens, $1.50 per million output tokens and $0.02 per million cached input tokens. For a limited time it was also free inside partner tools including GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode and Windsurf.

How well does it perform?

xAI reports 70.8% on the full SWE-Bench-Verified subset using its own internal harness. The company says benchmarks alone do not capture real software work, so it combines them with human reviews by experienced developers and automated behaviour checks, and reports that programmers rate the model fast and reliable for routine tasks.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Grok Code Fast 1 yet, so the facts on this page come from the sources listed above: x.ai/news/grok-code-fast-1. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.