It’s been another busy week in the xAI universe. Between a new coding model, an expanding developer platform, and the ongoing wait for Grok 5, there’s plenty to unpack. Here’s what happened.

Grok V9-Medium: A 1.5T Coding Model Arrives

The biggest news this week is Grok V9-Medium โ€” a 1.5 trillion parameter model purpose-built for coding, trained on Cursor data. Elon Musk confirmed it finished training and a public release is expected in mid-June 2026.

This is important for a few reasons:

  1. It’s not Grok 5. This is a specialized coding model, not a general-purpose flagship. xAI is taking the same path as OpenAI and Anthropic โ€” separate models for different use cases.
  2. Cursor training data is significant. Cursor is one of the most popular AI coding IDEs. Training on that data means Grok V9-Medium should understand how developers actually work โ€” multi-file refactors, debugging workflows, and agentic coding patterns.
  3. It’s a catch-up play. Musk publicly acknowledged xAI’s coding tools weren’t competitive with Claude Code or OpenAI Codex. This model is the direct response.

Grok Build 0.1: Developer Platform Goes Beta

Grok Build 0.1 hit public beta this week on the xAI API. This is xAI’s answer to Claude Code and Codex CLI โ€” a fast coding model with agentic capabilities, MCP (Model Context Protocol) support, and a developer-first experience.

Key features:

  • Plan Mode: Writes a full plan before touching code. You approve, comment, or rewrite before execution.
  • Sub-agents: Delegates work to specialized agents running in parallel across separate work trees.
  • Agent Dashboard: A new single-pane view for managing multiple coding sessions, blockers, and dispatches.
  • Plugin Marketplace: Ships with partner plugins from MongoDB, Vercel, Sentry, Chrome DevTools, Cloudflare, and Superpowers โ€” plus support for publishing your own.
  • Headless mode: The -P flag lets you wire Grok Build into CI/CD pipelines, cron jobs, and automation scripts via the Agent Client Protocol (ACP).
  • Composer 2.5: Available to SuperGrok and X Premium+ users, with stronger long-running task handling and instruction following.

xAI is clearly investing in developer tooling as a competitive moat. The marketplace approach is smart โ€” building a plugin ecosystem creates lock-in and community.

Grok 4.3 Skills: From Chatbot to Business Platform

xAI quietly launched Skills for Grok 4.3, and it’s actually a bigger deal than most coverage suggests.

Skills let users create persistent, reusable expertise across conversations. Built-in tools connect to SharePoint, Outlook, OneDrive, Google Workspace, Notion, GitHub, and Linear. You can also bring your own MCP server to build custom workflows.

What this means:

  • Grok is no longer just a chatbot โ€” it’s becoming a workflow hub
  • You can review code in GitHub, edit a Notion doc, and manage Linear tickets โ€” all from Grok
  • For enterprise IT, this is the kind of integration story that drives adoption
  • Skills are shareable and require no setup to use

This moves xAI from “AI model vendor” toward “AI platform” โ€” and it’s a direct shot at Microsoft Copilot’s enterprise integration narrative.

Grok Imagine Video 1.5: Topping the Charts

Grok Imagine Video 1.5 Preview is now ranked #1 on the Design Arena (Eloquent score 1357) and #2 in Artificial Analysis Video Arena, behind only ByteDance’s Seedance 2.0.

Specs:

  • 720p resolution, up to 15-second clips with native audio
  • Image-to-video with natural-language motion control
  • Average generation time: 41.2 seconds
  • Pricing: $0.01/image and $0.08/second for 480p; ~$8.40/minute for 720p with audio
  • A Quality Mode adds better text rendering and realism for enterprise use

For context, xAI claims it’s generating more images and video than all competitors combined โ€” a function of aggressive capacity deployment on Colossus infrastructure.

Grok Voice: Live on Mobile

Grok Voice rolled out June 4 โ€” conversational spoken interaction on the Grok mobile app. It’s a competitive response to ChatGPT’s voice mode and Claude’s voice features, and it brings Grok parity on the multimodal interaction front.

The Grok 5 Question

Here’s the elephant in the room: Grok 5 hasn’t shipped. The original Q1 2026 window came and went. The Q2 (May-June) window is now nearly closed. Polymarket gives it about a 33% chance of releasing by June 30.

What we know:

  • 6 trillion parameters, Mixture-of-Experts architecture
  • Training on the 1-gigawatt Colossus 2 supercluster in Memphis
  • Potentially expanding from 4-agent to 16+ agent system
  • Full API access may slip to Q3 2026

The delays aren’t necessarily bad news โ€” the current Grok 4.x models (4.3 for consumer, 4.20 Beta for API) are competitive. But with OpenAI, Anthropic, and Google all shipping fast, xAI can’t afford to miss another quarter without a new flagship.

Company Drama: Reorg, Departures, and a SpaceX Merger

The organizational story continues to be messy:

  • Only 2 of 11 original co-founders remain (Manuel Kroiss and Ross Nordeen)
  • Jack Garabedian (Starlink engineer) now runs the Grok training team, replacing original team
  • xAI reorganized into four divisions: Grok Main & Voice, Coding, Imagine (multimedia), and Macrohard
  • Macrohard / Digital Optimus โ€” the joint Tesla-xAI project for full enterprise automation โ€” is reportedly on pause
  • The SpaceX merger is done; IPO still ahead with combined valuation ~$1.25 trillion

For all the churn, xAI is still shipping product at an impressive pace. The question is whether the organizational instability catches up with the product velocity.

What to Watch

  1. Grok V9-Medium public release โ€” should happen any day now if mid-June holds
  2. Grok 5 โ€” will it ship by end of Q2? The window closes in 13 days
  3. OpenCode ecosystem growth โ€” integration with third-party agents could drive adoption
  4. xAI IPO โ€” likely later this year with SpaceX; the Grok 5 timeline will matter for investor sentiment

Stay tuned for next week’s update. If Grok 5 ships, we’ll have plenty to talk about.