SpaceXAI Grok 4.5 and the $60B Cursor Acquisition Reshape AI Coding

SpaceXAI — rebranded from xAI — launched Grok 4.5 in a two-week window from July 8 to July 16, 2026, alongside the finalized $60 billion acquisition of Cursor, the AI-powered code editor. Together, these announcements signal SpaceXAI’s transition from a research lab into an enterprise AI contender.

Grok 4.5: Intelligence per Unit Time

Grok 4.5 is SpaceXAI’s strongest model for coding, agentic tasks, and knowledge work. Trained alongside Cursor on tens of thousands of NVIDIA GB300 GPUs, it serves inference at 80 tokens per second — fast-model speeds that rival smaller, less capable models.

The model’s token efficiency is its standout metric. On SWE-Bench Pro, Grok 4.5 averages 15,954 output tokens per task, compared to Opus 4.8 at 67,020 — approximately 4.2 times fewer output tokens to solve the same problems. For teams paying per output token, this efficiency directly reduces costs.

Benchmark performance is competitive with frontier models. DeepSWE 1.0 scores 62%, close to Fable 5’s 66.1% and GPT-5.5 xhigh’s 64.31%. SWE Marathon resolution hits 29%, outperforming Opus 4.8 max (26%) and Fable 5 max (24%). Terminal Bench 2.1 reaches 83.3%, nearly matching Fable 5’s 84.3%.

Pricing and Positioning

Grok 4.5 pricing is set at $2 per million input tokens and $6 per million output tokens — highly competitive against Anthropic’s Sonnet 5 ($3/$15 standard) and Opus 4.8 ($5/$25). At 80 tokens per second, it also delivers faster responses than many comparably priced models.

SpaceXAI’s positioning centers on “highest intelligence per unit of time and cost” — a value proposition that resonates with cost-conscious engineering teams running large-scale AI coding pipelines.

The Cursor $60B Acquisition

The Cursor deal, which closed on June 16, 2026 at approximately $60 billion, gives SpaceXAI ownership of the most widely used AI-powered code editor. The strategic logic is clear: Grok 4.5 was co-trained on Cursor’s code-generation data, and the acquisition locks in that integration advantage.

With a $75 billion IPO on the horizon, SpaceXAI is building a vertically integrated AI coding stack: models (Grok 4.5), tools (Cursor), and infrastructure (the GB300 GPU cluster). This mirrors the full-stack approach OpenAI is pursuing with its own inference chip and Microsoft partnership.

Market Implications

The combined announcement reshapes the AI coding market. Three distinct ecosystems are emerging: OpenAI with GPT-5.6 and broad model family coverage, Anthropic with safety-first Sonnet/Opus models, and SpaceXAI with a tight coding-specific integration from model to editor.

For developers, the key takeaway is a third viable option for AI-assisted coding, with Grok 4.5’s token efficiency making it particularly attractive for high-volume agentic coding workflows.