SAN FRANCISCO — xAI on June 1 released Composer 2.5, a new agentic model now available inside Grok Build, the company's coding agent platform. Positioned as a fast, highly capable option for long-running and complex tasks, Composer 2.5 arrives with a substantially lower price point than its predecessor and expanded tool support.
Built for Agent-Scale Work
Composer 2.5 is specifically engineered for the demands of agentic workflows — the kind of long-horizon, multi-step tasks where most general-purpose models tend to lose coherence. xAI describes it as a fast, highly intelligent model with particular strengths in coding, JSON handling, tool use, function calling, and code execution.
The architecture has roots in Moonshot AI's Kimi K2.5 open-source checkpoint. xAI then applied 25 times more synthetic task training than went into Composer 2, the model it replaces — a significant investment aimed at improving instruction-following and task persistence over extended sequences.
Pricing and Access
Composer 2.5 starts at $0.50 per million input tokens and $2.50 per million output tokens. A faster variant is also available at $3.00 and $15.00 per million tokens respectively. Compared to Grok Build 0.1, the base tier represents approximately 100% cost reduction on input tokens — a meaningful drop for developers running high-throughput agentic pipelines.
The model is available immediately with no waitlist. Users can access it by navigating to build.grok.com and selecting Composer from the /models menu.
Cadence of Capability
Composer 2.5 lands as xAI continues shipping updates to Grok Build at an unusually fast pace. The platform debuted as a direct competitor to coding agents from Anthropic, Google, and OpenAI in late May. Elon Musk has publicly stated that "Grok Build is moving fast," and the Composer 2.5 release — just days after the platform's public launch — reflects that tempo.
For enterprise developers, the combination of lower per-token costs and stronger task-completion on agentic workloads positions Grok Build as a credible alternative to more established coding pipelines. The use of Kimi K2.5 as a base checkpoint is also a notable architectural choice, signaling xAI's willingness to build on open-source foundations when they offer a capability advantage.
What Comes Next
xAI's active model roadmap suggests Composer 2.5 is one step in a longer sequence. The company has seven model training runs active simultaneously on its Colossus 2 cluster — including Grok 5 variants at 6 trillion and 10 trillion parameters — meaning significant capability upgrades are already in the pipeline. Composer 2.5 reflects xAI's near-term posture: deploy capable, cost-efficient models for developers now while the flagship next-generation systems complete training. Developers building agentic pipelines today are positioning themselves to upgrade seamlessly as those larger models arrive.
