xAI Integrates Cursor's "Composer 2.5" into Grok Build
On June 1, 2026, xAI officially announced that Cursor's (Anysphere) coding-specialized model "Composer 2.5" is now available within its own terminal-based coding agent "Grok Build." It is offered as a fast model strong at long-running tasks and following complex instructions.
What Happened
The official xAI account (@xai) announced on June 1 that Composer 2.5 is now selectable within Grok Build, accompanied by a short demo video. The company positions the model as "fast, highly intelligent, and excellent at long-running tasks and following complex instructions." Announcement
Composer 2.5 is Cursor's in-house coding model officially released on May 18, 2026, and this time xAI has integrated it into its own product "Grok Build." Cursor official
Grok Build is a Rust-based coding agent CLI that xAI released in beta in mid-May 2026, featuring up to 8 parallel sub-agents, a "Plan mode" that reviews and approves plans before execution, and compatibility with existing plugins and workflows. It is offered to SuperGrok / X Premium Plus subscribers. xAI official
Background and Significance
Composer 2.5 is Cursor's in-house model built on a checkpoint of Moonshot's open-source model Kimi K2.5, with large-scale improvements via reinforcement learning (RL). It is said to have strengthened long-running tasks, following complex instructions, and natural collaboration through targeted textual feedback and a 25x scaling of synthetic tasks. It shows high scores on benchmarks such as CursorBench and is rated as delivering performance equal to or better than Claude Opus 4.7/4.8 or GPT-5.5 at a lower cost. Cursor official
What draws attention is that, in addition to its own models (such as grok-build-0.1), xAI has incorporated a powerful model from the competing ecosystem Cursor into Grok Build. This expands developer workflow choices, such as IDE (Cursor) versus CLI (Grok Build). There are also reports that Cursor is advancing next-generation model training utilizing the Colossus cluster, and a closer relationship with the xAI ecosystem has been noted. Analysis article
Grok Build is positioned to compete with established coding agents such as Claude Code (Anthropic) and Cursor. DevOps.com
In spec comparisons, Composer 2.5 is said on some comparison sites to have a 1M-token context window and to be strong at iterative editing, while Grok Build 0.1, with 256K tokens, is said to be strong at autonomous greenfield development. Cursor's reference pricing is standard input $0.50/M, output $2.50/M; the Fast variant is input $3.00/M, output $15.00/M. The billing scheme within Grok Build depends on xAI's subscription, and details are undisclosed. Comparison site
Reactions
On X, developer voices rose actively right after the announcement. Positive feedback dominated, with comments such as "more useful than Opus 4.8 or GPT-5.5," "the speed is insane—a 6-minute GPT-5.5 task in under 1.5 minutes," "blazing fast with parallel sub-agents," "best cost performance compared to Claude/GPT," and "especially strong on long-running tasks." Reactions
In comparisons with Grok Build's own model, voices such as "Composer 2.5 has the edge" and "Grok Build 0.1 makes many tool calls and consumes credits heavily, but Composer is still better" were also raised. Cursor forum
On the other hand, a minority pointed out caveats such as "it still struggles with production-grade Python/Rust generation on large codebases," "the difference between the fast and standard versions is unclear," and "usage can spike easily." Overall, its price-performance ratio and durability on long-running tasks are especially appreciated. Note