xAI launches Grok Composer 2.5 inside Grok Build, trained on 25x more synthetic tasks

xAI announced Composer 2.5 on June 1, positioning it as a state-of-the-art model within the Grok Build environment aimed squarely at agentic developer workloads. The model targets long-running tasks and complex instruction-following, with specific strengths in coding, agents, structured JSON output, and tool use — including function calling and direct code execution, the primitives that matter for autonomous coding agents.

The headline technical claim is scale of training signal: Composer 2.5 was trained with 25 times more synthetic tasks than its predecessor, Composer 2. The heavy reliance on synthetic task generation mirrors an industry-wide trend of bootstrapping agentic capability through self-generated training data rather than scarce human-labeled examples. Access is gated to SuperGrok and X Premium+ subscribers via the /models menu in Grok Build.

The launch puts xAI into the same agentic-coding arena as Anthropic's Claude Code, OpenAI's Codex (now on Bedrock), and Alibaba's Qwen3.7-Max, intensifying a coding-agent race that has become the clearest commercial battleground for frontier models. xAI's distribution advantage is X's subscriber base and Elon Musk's reach.

The single-source nature of the launch (a third-party blog) means independent benchmarks are scarce so far, so the "state-of-the-art" claim should be treated cautiously until external evals land. Watch for head-to-head coding-benchmark results against GPT-5.5 and Claude Opus 4.8, and whether xAI exposes Composer 2.5 via API beyond the Grok Build subscriber gate.