Grok 4.5: SpaceX Paid $60B for Cursor and Built This Model

xAI launched Grok 4.5 on July 8, 2026, at $2 per million input tokens and $6 per million output tokens — 60% cheaper than Claude Opus 4.8. Six weeks earlier, SpaceX had acquired Cursor, the most widely used AI coding editor, for $60 billion in stock. Grok 4.5 is the first model to come out of that combined company.

Elon Musk described it as "an Opus-class model, but faster, more token-efficient and lower cost." The benchmark numbers support that framing on specific dimensions. On the dimensions where they do not, the pricing difference is significant enough to matter.

What Happened: SpaceX Bought Cursor for $60B

The Cursor acquisition is the context everything else flows from. SpaceX announced the deal on June 16, 2026, four days after going public at $135 per share in the largest IPO in history, raising $75 billion.

Cursor had been one of the fastest-growing developer tools in the world — a VS Code fork that put Claude, GPT-4o, and other models directly inside the coding environment. By mid-2026, it had become the default coding interface for hundreds of thousands of engineers. SpaceX paid $60 billion in stock to own that distribution.

Musk had already merged SpaceX with xAI earlier in 2026. The combined company — sometimes referred to as SpaceXAI — now controls rocket manufacturing, satellite internet, the Grok family of AI models, and the world's most popular AI coding editor. Grok 4.5 is the first model built to run natively inside Cursor.

The deal is pending regulatory approval and expected to close in Q3 2026. Grok 4.5 launched ahead of formal close, which means Cursor users are not automatically on Grok 4.5 yet — but the model is available via API and in xAI's own interfaces.

Grok 4.5 Specifications

Grok 4.5 ships with a 500,000-token context window, configurable reasoning effort, and an OpenAI-compatible API. The reasoning effort parameter lets developers dial between fast responses and extended thinking on a per-request basis — similar to how Claude Sonnet 5 handles reasoning modes.

The model was trained across tens of thousands of NVIDIA GB300 GPUs. xAI has not published a parameter count.

Pricing:

• Input: $2 per million tokens
• Output: $6 per million tokens

For comparison: Claude Opus 4.8 is $15 input / $75 output. Claude Sonnet 5 at introductory pricing is $2 input / $10 output. Grok 4.5 undercuts Sonnet 5 on output by 40% and matches it on input.

Benchmark Results

On Terminal-Bench 2.1, Grok 4.5 scores 83.3% versus GPT-5.6 Sol at 88.8%. That is a 5.5-point gap on the benchmark that matters most for agentic coding. On SWE Bench Pro, a real-world software engineering benchmark, Grok 4.5 resolves 64.7% of issues.

The Artificial Analysis Intelligence Index places Grok 4.5 fourth, above every open-weight model and above all Gemini models. The positioning — below Sol but ahead of the rest — combined with the $2/$6 pricing is the core commercial pitch.

Our Analysis: Who Should Use Grok 4.5

The model is competitive on coding benchmarks but not the frontier leader. GPT-5.6 Sol and Kimi K3 both outperform it on Terminal-Bench 2.1. Claude Sonnet 5 has not been fully benchmarked on Terminal-Bench yet, but its introductory pricing at $2/$10 is higher on output than Grok 4.5's $6.

The scenario where Grok 4.5 makes clear sense is high-volume coding pipelines where you need Opus-class quality at lower cost than Sonnet 5. If your workload is SWE-Bench-style issue resolution — finding and fixing bugs in existing code, implementing features from specifications — the 64.7% SWE Bench Pro score places Grok 4.5 in competitive territory at a price point that is hard to match.

Where it makes less sense: frontier reasoning tasks (GPQA-type problems), extended context analysis, or any workload where you need the absolute highest score on Terminal-Bench. For those, GPT-5.6 Sol or Kimi K3 (when weights drop July 27) are the better calls.

The Cursor integration is the variable to watch. When the SpaceX-Cursor deal closes in Q3, Grok 4.5 will have direct access to the editing context, file tree, and codebase state that Cursor already tracks. That integration, if done well, has the potential to significantly outperform API-based coding models that lack editor context — even ones with better standalone benchmarks.

What the SpaceX-Cursor Deal Actually Means for Developers

Cursor was a neutral tool. It ran whichever model the developer selected — Anthropic, OpenAI, Google, or open-weight alternatives. After the SpaceX acquisition, it is no longer neutral. It is part of an AI company that has its own flagship model.

This does not mean Cursor will block other models. The business case for Cursor depends on developers trusting it as a multi-model interface. Forcing Grok 4.5 on users would accelerate migration to Windsurf, Void, or other editors. xAI will make Grok 4.5 the default or the promoted option, not the only option.

What changes is the incentive structure. When Cursor surfaces model recommendations, Grok 4.5 will be the house model. When Cursor-specific features ship — context-aware completions, deeper file integration — they will optimize for Grok 4.5 first. Other models will be second-class citizens in their own right, not by being blocked, but by being behind on integration depth.

Developers who care about editor lock-in should be paying attention now, not after the deal closes.

Key Takeaways

• Grok 4.5 launched July 8, 2026 at $2/$6 per million input/output tokens — 60% cheaper than Claude Opus 4.8 at $15/$75
• SpaceX acquired Cursor for $60B on June 16, 2026 — the largest acquisition in AI coding tools; deal closes Q3 2026
• 500K-token context window, configurable reasoning effort, OpenAI-compatible API
• Terminal-Bench 2.1: 83.3% — 5.5 points below GPT-5.6 Sol (88.8%); SWE Bench Pro: 64.7%
• 4th on Artificial Analysis Intelligence Index — above all open-weight models and Gemini, below Sol and Kimi K3
• The Cursor integration is the real bet — editor context awareness will matter more than standalone benchmarks once the deal closes
• What to watch: Q3 Cursor integration depth, and whether xAI makes Grok 4.5 the default model in Cursor post-acquisition