Update, October 9, 2026: Cursor’s release post is dated July 8. xAI confirmed API access on July 16.

What happened

Cursor announced Grok 4.5 on July 8, with availability across its desktop, web, iOS, CLI and SDK products. xAI’s July 16 release post subsequently confirmed API availability for the model, which it describes as built for coding, agentic tasks and knowledge work. Cursor also said individual and team plans included first-party pool usage with extra access during the first week.

xAI listed a base API price of $2 per million input tokens and $6 per million output tokens at launch. It also reported benchmark results and lower output-token use on software tasks.

Why it matters

Grok 4.5 is not being introduced as a narrow coding specialist. Both announcements describe a model that can work across software, research, data, and other computer-based tasks. For builders, that points toward a workflow where the same agent can investigate a problem, change code, use tools, and verify the result over a longer task.

The Cursor release also makes the distribution story more important than the model name. Builders can try Grok 4.5 inside a tool they may already use, without setting up a separate API integration first. At the same time, the model’s API pricing gives builders a way to evaluate it directly in their own app or agent workflow.

Cursor disclosed an important limitation in its benchmark notes: an earlier snapshot of its codebase was accidentally included in training, which may have advantaged Grok 4.5 on CursorBench. That makes the benchmark comparison less useful than a focused test on representative tasks.

Who should care

Cursor users working on multi-file changes, difficult bugs, or tasks that require repeated tool use should pay attention. Builders choosing an API model for coding or agentic workflows should also compare Grok 4.5 against their current default on quality, latency, output-token use, and total cost.

What builders should do next

Try Grok 4.5 on one task that regularly exposes the limits of your current model: a cross-file refactor, a bug that requires investigation, or a small feature that needs tests and verification. Keep the prompt and codebase constant, then compare the result rather than relying on the leaderboard.

If you use Cursor, start there before building a direct API integration. If you call the API directly, treat the advertised price as only one part of the calculation. Long agent runs can still become expensive if the model needs many turns or returns large amounts of code.

End of article