The post-SpaceXAI consolidation era continues to accelerate. This week, xAI’s new parent SpaceXAI shipped the Grok Build 0.1 coding model on the public API, opened Grok authentication inside Kilo Code, and confirmed that the next-generation V9-Medium foundation model — three times larger than the current production model — has completed training. Add a batch of platform API upgrades (Context Compaction, WebSocket mode, Smart Turn STT, Image Search) and SpaceX is making a clear statement: the agentic infrastructure is no longer a side project. Here is the xAI Weekly for June 3, 2026.
Grok Build 0.1 Now Available on the API
On May 29, xAI announced that Grok Build 0.1 — the dedicated coding model that powers the Grok Build CLI — is now available via the xAI API in public beta.
The model slug is grok-build-0.1 and it is specifically trained for agentic coding tasks: web development, debugging, MCP support, and multi-step software engineering workflows. Key specs:
- 256K token context window
- $1.00 / 1M input tokens, $3.00 / 1M output tokens
- Supports text and image inputs with text output
- Optimized for interactive coding agents, tool use, and autonomous task loops
The API launch follows the Grok Build CLI beta that opened to all SuperGrok subscribers on May 25 (previously gated to SuperGrok Heavy / $299/month). xAI also introduced a $99/month SuperHeavy promotional tier for the first six months — a significant reduction from the original Heavy pricing.
(docs.x.ai release notes, x.ai news)
Why It Matters for Engineering Teams
Grok Build 0.1 on the API means teams can now build custom orchestrators, CI/CD integration pipelines, and automated code review agents on top of the same model that powers the CLI — without requiring every developer to run the terminal client. For organizations evaluating coding agents, this opens up integration patterns that the CLI-only approach didn’t support.
Grok Build remains one of three credible coding agent options alongside Claude Code and Codex CLI — our enterprise coding agent harness guide provides a framework for evaluating them side by side.
Grok Integrates with Kilo Code via OAuth
On May 27, xAI announced that SuperGrok and X Premium+ subscribers can now use Grok directly inside Kilo Code — the open-source agentic coding platform that supports VS Code, JetBrains IDEs, and the terminal.
The integration uses OAuth-based authentication, meaning developers connect their Grok subscription with no separate API key management. No API key to provision, rotate, or leak — just a sign-in flow. This matters for enterprise security teams managing credential sprawl across multiple AI coding tools.
Kilo Code already supports Claude Code, Codex, and Cursor in the same interface. Adding Grok to the roster makes it a genuinely multi-model agentic IDE — lowering the friction for teams that want to compare or switch between providers without changing their development environment.
(x.ai news: Use Grok in Kilo Code)
V9-Medium: 1.5 Trillion Parameters, Mid-June Release
On May 25, Elon Musk announced that Grok V9-Medium has completed training — a 1.5 trillion-parameter model, three times larger than the current v8-small production model (approximately 500B parameters). Supervised fine-tuning is underway, with reinforcement learning beginning shortly and a public release expected in mid-June 2026.
Cursor Training Data
The most consequential technical detail is the training data. V9-Medium was trained on Cursor workflow data — real-world developer sessions from the AI-augmented code editor — not just public GitHub repositories. This follows SpaceXAI’s April agreement with Anysphere (Cursor’s developer) that includes an option to acquire the company for $60 billion. The Cursor data pipeline gives V9-Medium visibility into how actual engineers debug, refactor, and ship code — a fundamentally different training corpus from the public-repo-centric approach most coding models use.
Competitive Positioning
The current Grok 4 (v8-small) scores approximately 75% on SWE-Bench Verified, compared to Claude Opus 4.7 at 87.6%. Whether tripling parameter count and incorporating Cursor workflow data closes that gap is what mid-June will test. For organizations currently evaluating the xAI ecosystem, our xAI and Grok enterprise buyer’s guide covers vendor assessment, pricing comparison, and migration planning. Analysts caution that raw parameter scale doesn’t guarantee proportional performance gains — Mixture-of-Experts architectures (like DeepSeek’s) can match dense models at a fraction of active parameters — but the Cursor data strategy suggests SpaceXAI is focused on data quality, not just scale.
Open-Source Commitment
Musk confirmed the current v8-small model (500B params) is planned for open-source release by end of 2026, continuing the pattern: Grok 1 (Apache-2.0, March 2024), Grok 2.5 (community license, August 2025), and Grok 3 (pledged February 2026). For the open-source AI community, a 500B-parameter model from a frontier lab represents a substantial foundation model — once it arrives.
Platform API Upgrades
The xAI May release notes include several developer-facing improvements shipped over the past week:
Context Compaction API
New endpoint that shrinks long conversation histories into a shorter compressed context for follow-up requests. Designed to reduce cost and time-to-first-token on long agent loops — particularly relevant for teams running multi-turn agent orchestration. (Context Compaction docs)
WebSocket Responses API Mode
A new WebSocket mode for the Responses API that maintains a single long-lived connection for tool-heavy agent workloads, reducing end-to-end latency by eliminating connection setup overhead per turn. (WebSocket Mode docs)
Smart Turn for Streaming STT
The Speech-to-Text API now supports ML-based end-of-turn detection via the smart_turn parameter — reducing false endpointing during dictation, number sequences, and mid-sentence pauses. Practical for voice agents and real-time transcription use cases.
Image Search in Web Search Tool
The Web Search tool now supports explicit image search. Enable enable_image_search to let Grok find and embed relevant images in responses — useful for content generation, research, and presentation workflows.
What to Watch
- Grok Build 0.1 API adoption: The public beta pricing ($1/$3 per 1M tokens) undercuts most dedicated coding models. Watch for adoption patterns among CI/CD pipeline builders and agent orchestrator developers.
- V9-Medium release (mid-June): The 1.5T parameter model with Cursor training data is the most consequential SpaceXAI product launch since the merger. SWE-Bench scores at release will define whether the parameter scaling strategy pays off.
- Kilo Code multi-model surface: Grok joining an IDE that already supports Claude Code, Codex, and Cursor creates a true multi-model switching environment. Enterprise teams evaluating coding agents now have a single integration point for head-to-head comparisons.
- Context Compaction for agent economics: The economic impact of context compaction on long-running agent loops could be significant for production deployments. Watch for third-party benchmarks comparing token savings.
- v8-small open-source timeline: End-of-2026 open-source release of a 500B-parameter frontier model would be one of the largest publicly available model releases to date.
That is this week’s xAI Weekly. The Grok Build 0.1 API launch and Kilo Code integration signal xAI’s transition from a consumer chatbot company to a full-stack infrastructure provider for agentic coding. V9-Medium’s mid-June release will be the defining moment — and the open-source commitment to v8-small keeps SpaceXAI in the developer community’s good graces.
If your team is evaluating Grok Build for coding workflows, planning a Kilo Code deployment, or navigating the agentic coding tool landscape, book a strategy session with Big Hat Group — we help enterprise teams make sense of this rapidly consolidating market.
Check back next week for the latest.