xAI launches Grok Composer 2.5, a coding agent built on Kimi K2.5
25x more synthetic training data, 100% cheaper input tokens than Grok Build 0.1.
xAI has announced Grok Composer 2.5, a new agentic coding model now available inside Grok Build. According to an xAI post on June 1, the model is designed for long-running, complex instruction-following tasks where most general-purpose models falter. It is built on the open-source checkpoint of Moonshot's Kimi K2.5 and trained with 25 times more synthetic tasks than its predecessor, Composer 2. Key strengths include coding, agentic workflows, JSON generation, and tool use such as function calling and code execution.
Pricing is set at $0.50 per million input tokens and $2.50 per million output tokens. A faster variant costs $3.00 and $15.00 per million tokens, respectively. xAI claims the input token price is roughly 100% cheaper than Grok Build 0.1. Developers can access the model immediately at build.grok.com by selecting Composer from the /models menu — no waitlist required. This release signals xAI's continued push into specialized coding agents for professional use.
- Built on Moonshot's Kimi K2.5 open-source checkpoint with 25x more synthetic training data than Composer 2.
- Pricing: $0.50/M input tokens, $2.50/M output tokens; faster variant at $3/$15 — 100% cheaper on input vs Grok Build 0.1.
- Optimized for long-running tasks, coding, agents, JSON, tool use, function calling, and code execution.
Why It Matters
Grok Composer 2.5 offers a cost-effective, specialized coding agent for complex, multi-step tasks in production environments.