Grok 4.7 rolls out to Grok web and mobile apps

The updated model began replacing the previous release across Fast, Expert, Build, and Heavy modes in a staged rollout started on October 1.

On October 1, 2026, Grok 4.7 began rolling out to Grok web and mobile apps as the base model across Fast, Expert, Build, and Heavy modes. Product tracking accounts TestingCatalog and LuminaBench first flagged the release using interface screenshots, and the official @grok account on X confirmed the rollout the following day. The change delivers the model to general chat users ten days after SpaceXAI initially released it only to software developers and coding platforms.

How Grok 4.7 powers app modes

Inside Grok apps, users choose different modes based on the task at hand. Fast provides quicker replies, Expert triggers deeper reasoning (labeled “Thinks hard”), Heavy engages multiple agents (“Team of Experts”), and Build focuses on writing software. Instead of maintaining separate models for each option, SpaceXAI uses Grok 4.7 as the shared foundation, adjusting reasoning effort and compute infrastructure behind each mode. Chat users interact with the newer model directly without configuring technical parameters.

From the developer API to consumer chat

SpaceXAI announced Grok 4.7 on September 21, 2026, limiting initial access to the Cursor editor, the Grok Build environment, and the xAI programming interface (API) at $2 per million input tokens and $6 per million output tokens. A token is a basic text unit processed by the model. The ten-day gap between developer availability and the consumer app release drew comments from regular users on social media. Technical documentation updated on October 5 lists a context window of 500,000 tokens, a knowledge cutoff of May 2026, and native text and image input.

Performance according to the company

In its launch announcement, SpaceXAI published internal benchmark numbers comparing Grok 4.7 at high effort against Grok 4.6. On CursorBench 4.0, which evaluates coding inside an editor, the model scored 46.3% compared to 40.4%. On Terminal-Bench 4.0, which tests command-line problem solving, scores rose from 20.3% to 37.6%. In the electrical engineering benchmark EEBench, Grok 4.7 reached 64.0% against 53.0%, while DeepSWE v1.1 reached 71.0% under high reasoning effort, up from 65.2%. All figures were measured internally by SpaceXAI rather than by independent evaluation labs.

Staged rollouts and documentation gaps

The update rolls out in phases and did not reach all accounts immediately. Free accounts and the Grok module on X continued to surface Grok 4.6 or 4.5 during the opening days of October. In addition, the consumer rollout arrived without a dedicated post on the x.ai/news blog or mentions in developer release notes. The public pricing page on x.ai also continued listing Grok 4.6 for SuperGrok subscribers well after the update began. On the same day, SpaceXAI separately introduced Primary Bot, a proactive suggestions tool that operates independently from the chat model update.

This newsroom has not tested the app modes. SpaceXAI sells Grok subscriptions and API access.