Video: "AI Agents Just Changed Forever: GLM 5.2 The Best Open-Source Model Ever?" by Julian Goldie on YouTube.
What SpaceX buying Cursor actually means
SpaceX announced the $60 billion all-stock deal on 16 June 2026, just days after its IPO. Cursor — the AI coding assistant built by Anysphere — had roughly $2.6 billion in annualised revenue at the point of acquisition and was last valued at $29.3 billion in late 2025. The deal is expected to close by the end of September 2026.
The strategic logic: SpaceX has been quietly building an AI division — absorbing xAI earlier in the year — and now wants a direct coding tool to compete with OpenAI Codex and Anthropic's Claude Code. The joint model that SpaceX and Cursor have apparently been training together will appear inside both Cursor and Grok Build once the deal closes.
For UK businesses using Cursor today, the short-term practical impact is minimal. The product keeps working, the team stays in place, and the merger needs regulatory sign-off before anything changes structurally. The longer-term question is whether Cursor shifts from a developer-focused product toward something more tightly integrated with Elon Musk's wider AI stack.
GLM 5.2 and why the open-weight rankings matter
Z.ai's GLM 5.2 launched on 13 June 2026 with a full MIT licence, a 1 million token context window, and IndexShare sparse attention — a memory technique that keeps large-context tasks cheaper to run. It topped Artificial Analysis's open-weight model leaderboard shortly after launch, ahead of the previous leaders from Meta and Mistral.
That benchmark position matters for a specific reason: GLM 5.2 can be downloaded and run locally, without routing prompts to a commercial API. For businesses handling commercially sensitive data — client records, unreleased products, financial figures — local inference means that information does not leave the building. The MIT licence also means it can be used in commercial products without royalty concerns.
Julian tested GLM 5.2 on creative coding tasks in an earlier session and found it won four of five tests against Claude Opus 4.8 and Kimi K2.7, including a liquid simulation, a labelled interactive solar system, and a neon arcade game. Those are difficult, multi-step tasks, not simple one-liners. The model is genuinely capable, not just well-positioned on a benchmark chart.
Why these two stories sit together
Julian framed the week as a shift in who controls the AI coding stack. When a proprietary tool like Cursor gets absorbed by a large platform company, its roadmap shifts to serve the parent organisation's goals. That is not automatically bad — more resources can mean faster development — but it does mean the product's priorities may diverge from what developers actually wanted.
A top-performing open-source alternative like GLM 5.2 available in the same week is significant. Teams that want predictability, local execution, and no dependency on a platform company's appetite now have a credible option. Worth knowing: running a model locally still requires capable hardware, and setup takes more than a few minutes. It is a genuine trade-off, not a direct replacement for a polished product like Cursor.
What's overhyped versus what's genuinely useful
The "AI agents changed forever" framing in Julian's title is typical YouTube energy — this week's news is significant without being a rupture. Cursor will not change overnight; GLM 5.2 is excellent but still needs configuration to do what a managed service does out of the box. The useful takeaway is simpler: there are now two credible paths for AI-assisted coding — a well-funded proprietary tool inside a large platform, or a free open-weight model you run yourself. The right one depends on what your team does and what your risk tolerance is.
Where this connects to NordSys
If tracking this kind of AI-agent news makes you wonder whether one could actually run inside your business, that's exactly what our AI Agents do — named, briefed and managed for you, no setup fee, from £6 a day.
See our AI Agents →