Grok 4.5 In Cursor: A Coding Model With Competitive Ambitions
Grok 4.5 is a large language model released by SpaceXAI and tightly integrated into Cursor’s AI coding environment, designed to handle long‑context software development, agentic workflows, and complex knowledge work while competing directly with leading models used for AI code completion and enterprise coding tasks. The headline story is not another chatbot but a model built to live inside the editor, drive agents across real codebases, and challenge incumbents like Opus and GPT in the environments where developers spend their time. SpaceXAI, formed after SpaceX absorbed xAI, has released Grok 4.5 as its first jointly built model with Cursor and its first major launch since going public. Cursor has already rolled out support, making Grok 4.5 a core engine for Cursor AI coding across desktop, web, mobile, CLI, and SDK. In effect, this partnership turns Cursor from a neutral client into a channel for SpaceXAI’s ambitions in developer tooling.
A Jointly Trained Grok 4.5 Model, Tuned On Real Developer Work
The most important shift is how Grok 4.5 was built: as a mixture‑of‑experts system trained jointly by SpaceXAI and Cursor on trillions of tokens from real Cursor user sessions. That means the Grok 4.5 model has been shaped not only by public code and text, but by concrete patterns of how coding agents explore codebases, respond to failures, and iterate through tasks. Cursor says SpaceXAI ran training across tens of thousands of Nvidia GB300 GPUs while applying aggressive data filtering, deduplication, and domain‑focused curation. This is the first joint development effort between the two, and it shows a clear intent: align the model’s behavior with how developers actually work rather than with generic chat. Cursor also broadened the training beyond strict software engineering to include STEM tasks, research papers, finance, and legal work so Grok 4.5 can support teams whose coding sits inside wider analytical workflows."Grok 4.5 was trained in close collaboration with Cursor on real developer usage," as one source puts it.
Performance, Pricing And What They Mean For AI Code Completion
On benchmarks, Grok 4.5 does not dominate across the board, but it lands in the same tier as the current leaders while undercutting some of them on price. On Terminal‑Bench 2.1, it reaches 83.3%, edging out Opus 4.8’s 78.9% and coming within a fraction of GPT‑5.5’s 83.4%. It also beats both Opus 4.8 and GPT‑5.5 on DeepSWE 1.0, though it trails Opus 4.8 on SWE‑Bench Multilingual and SWE‑Bench Pro and is behind Fable 5 on all shown benchmarks. The more interesting story is efficiency: on SWE‑Bench Pro tasks, Grok 4.5 uses about 4.2 times fewer output tokens than Opus 4.8 while running at roughly 80 tokens per second. For teams doing iterative AI code completion and refactoring, fewer tokens per task mean lower cost and faster turns. Through Cursor and Grok Build, Grok 4.5 is priced at USD 2 (approx. RM9.2) per million input tokens and USD 6 (approx. RM27.6) per million output tokens, below Opus 4.8’s USD 5 (approx. RM23) input and USD 25 (approx. RM115) output, and in a similar band to GPT 5.6 Luna’s pricing."On SWE‑Bench Pro tasks, Grok 4.5 uses roughly 4.2 times fewer output tokens than Opus 4.8," one benchmark summary notes.
Cursor AI Coding Becomes A SpaceXAI Channel For Enterprise Workflows
The integration changes Cursor’s role in the AI coding landscape. Grok 4.5 is now a core engine inside Cursor’s AI coding environment, giving developers access to a high‑capacity model directly within their editor and workflows. The model supports both text and image input and is built for long‑context coding sessions with a context window of up to 500,000 tokens. That makes Cursor a natural front end for agentic work on large monorepos, complex data pipelines, and cross‑functional projects that blend code with financial or legal analysis. At the same time, Cursor keeps its lighter Composer 2.5 model available for users who prefer a smaller specialist option. For enterprise developers, the message is clear: there is now a serious alternative to established AI coding tools that matches or exceeds top‑tier models on some coding benchmarks while improving speed and output‑token efficiency. The release arrives in the same week that GPT‑5.6 is rolling out more broadly, intensifying a three‑way fight over which assistant will own the modern coding workflow.
Strategic Timing And The Road Ahead For Grok 4.5
SpaceXAI’s Grok 4.5 release is not an isolated event but the first visible outcome of Elon Musk’s decision to overhaul xAI’s leadership, absorb it into SpaceX, and acquire Cursor in a USD 60 billion (approx. RM276 billion) deal announced in mid‑June. It is also SpaceXAI’s first major model launch since going public. The timing, alongside GPT‑5.6’s broader rollout, signals that SpaceXAI does not intend to cede the developer tooling market to OpenAI and Anthropic. Practically, Grok 4.5 is live today in Grok Build, across all Cursor plans, and via the SpaceXAI console using an API key. It is not yet available in the EU, with access expected to open in mid‑July and EU availability later in July 2026. Cursor has also removed accidental training contamination from an earlier snapshot of its own codebase in preparation for future models. The direction of travel is clear: this partnership aims to turn Cursor into the default home for SpaceXAI releases, targeting enterprise developers who want powerful AI code completion and agentic systems without locking themselves into the existing giants.
