Cursor has unveiled Grok 4.5, an “Opus-class” coding and agentic model built jointly with SpaceXAI — the entity formed after xAI merged into SpaceX earlier this year — and made it available across Cursor’s desktop, web, iOS, CLI, and SDK surfaces. The headline is pricing: $2 per million input tokens and $6 per million output tokens for the base model, with a faster variant at $4/$18. That’s roughly half of what teams have been paying for comparable frontier coding models over the past year, and it lands the same week GitHub folded its first open-weight model into Copilot.
We’ve tracked steady price and positioning shifts across the AI coding tool market all year — Cursor’s own move to usage-based enterprise pricing, GitHub Copilot’s billing changes, Windsurf’s rebrand into Devin Desktop. Grok 4.5 is a different kind of move: it’s not a plan restructure, it’s a direct price cut on frontier capability, aimed squarely at Anthropic and OpenAI’s coding-model pricing. When a well-capitalised entrant undercuts on price rather than just competing on benchmark scores, it’s usually a signal that the market is maturing from “which model is smartest” into “which model is smartest per dollar” — a much more commercial, much more contestable question.
So what
If your team is choosing or re-negotiating an AI coding tool stack, price-per-token is now a live variable worth revisiting quarterly rather than locking in once and forgetting about it. It doesn’t change the fundamentals of what makes a build production-ready — model choice affects velocity, not architecture, security, or maintainability — but it does mean the cost side of “build with AI assistance” is shifting fast enough that a stack chosen six months ago may no longer be the best-value one. If you want a second opinion on tooling choices as part of a build, our AI-assisted development work covers exactly this, or get in touch to talk through your options.