Visit NinjaChat: http://ninjachat.ai/
Visit ByteRover: https://www.byterover.dev/?source=ack6
In this video, I’ll be telling you about xAI’s new Grok Code Fast model—its pricing, speed, strengths and limitations, how it compares to Deepseek and GPT-5 Mini, where you can try it for free, and how to plug it into Claude Code via an Anthropic-compatible API.
—
Key Takeaways:
🚀 xAI launches Grok Code Fast, a first-gen coding-focused model that’s effectively the same as Sonic.
💸 Priced at $0.20 and $1.50 respectively, making it one of the cheapest coding models.
⚡ Around 80 tokens per second—faster than Deepseek in my tests.
🧪 Benchmarks matched Sonic with no noticeable improvements on my runs.
🧠 Limited reasoning (1–2 lines) and can go off-track; not ideal for complex tasks, UI, or architecture.
🧰 Strong tool-calling and great for small, iterative edits; pair it with Deepseek as the architect.
🌐 Try it free via KiloCode until the weekend (no limits), also available on RooCode and OpenCode.
🔗 Works with Claude Code through xAI’s Anthropic-compatible API; route via Requesty or OpenRouter, and add ByteRover (MCP) for persistent memory.
✅ For most coding, GPT-5 Mini or Deepseek V3.1 are still better/cheaper, with adjustable reasoning and verbosity.
—
Timestamps:
00:00 – Introduction
00:07 – What is Grok Code Fast?
00:45 – Pricing and speed
01:22 – Benchmarks vs Sonic & Strengths
02:23 – Where to use it free: KiloCode, RooCode, OpenCode
04:02 – Using it in Claude Code (Anthropic-compatible API)
04:18 – NinjaChat (Sponsor)
05:07 – Requesty/OpenRouter routing
06:05 – Memory, context, and ByteRover (MCP)
08:03 – Final thoughts and next steps
source
