Grok Code Fast 1
2025-08-28 · Publisher : xAI · Domain : Artificial Intelligence
xAI launches grok-code-fast-1, a fast and economical reasoning model designed for agentic coding, available in the API and free for a limited time from several partners.
What changes
- New architecture, a pretraining corpus rich in code and post-training on real pull requests.
- Good command of common tools (grep, terminal, file editing), with cache rates claimed above 90% at partners.
- Languages cited: TypeScript, Python, Java, Rust, C++ and Go.
- Temporary free access via GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode and Windsurf; the model had been circulating in testing under the codename "sonic".
### Specifications - API identifier: grok-code-fast-1. - Claimed speed: about 190 tokens per second. - Context and knowledge cutoff: not disclosed in the blog post. - Availability: xAI API and partners. A multimodal variant with parallel tool calls and a longer context was in training. ### Pricing Per million tokens, in USD. | Input | Output | Cached input | |---|---|---| | $0.20 | $1.50 | $0.02 | ### Benchmarks Score published by the vendor: 70.8% on SWE-Bench-Verified (full subset, xAI's internal harness). xAI notes that this benchmark poorly reflects real-world usage. ### Key takeaways A low-cost, highly responsive coding model, designed for agent loops in IDEs.
Links
Tags : LLM, xAI, Grok, Grok Code, Codage