Following its acquisition of AI coding platform Cursor, SpaceXAI, the rebranded AI division of Elon Musk’s SpaceX, has launched Grok 4.5 in a move aimed at the enterprise market.
Introduced on July 8, Grok 4.5 is built for coding, agentic tasks and knowledge work. The model was trained alongside Cursor, the vendor said. The training was on the Nvidia GB300 platform, which combines GPUs and CPUs. SpaceXAI said it invested heavily in data filtering and curation, and scaled reinforcement learning with a focus on per-token intelligence.
The new model is SpaceXAI’s first model specifically designed for agentic coding. Previously, the vendor, known as xAI before being renamed SpaceXAI on July 6, ventured into coding with its Build CLI, a coding agent for developers. However, the Grok model has long been aimed mostly at consumers using Musk’s X social media platform, with little enterprise adoption.
Grok 4.5 shows how SpaceXAI is targeting the enterprise market, aiming to rival AI coding models such as Claude Code from Anthropic and Codex from OpenAI, which are regarded as the preferred AI coding platforms. The model release comes on the cusp of SpaceX’s $75 billion IPO and its June 16 acquisition of the popular agentic coding platform Cursor for $60 billion in stock.
“The idea, I guess, is to continue to push the popularity of Cursor among developers,” said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget.
SpaceXAI trained Grok 4.5 on trillions of tokens of Cursor data, which enables the model to run with both existing software and in developer-agent interactions. Cursor and SpaceXAI revealed that, although Grok 4.5 is optimized for coding, its training data is somewhat broader than that of Cursor’s coding model, Composer 2.5. Grok 4.5 and Cursor models are meant to be complementary.
A Shift in Platform
In addition, SpaceXAI is showing that Grok is more than a model that X users can interact with on the platform.
“They’re coming out from under the shell of being fully associated with Twitter/X,” said Bradley Shimmin, an analyst at Futurum Group. “That is a good thing for them … You can have more options to get to the Grok models, and I hope that continues going forward.”
The vendor’s focus on coding is also good for the market because it leads to more competition, which benefits enterprises the most, Shimmin said. He added that competition and an array of models produce greater optionality, enabling enterprises to evaluate what differentiates each model, the platform and tooling associated with them, and the user experience that comes with each model.
“If you’re a practitioner, the only way to fully gauge the value of a new model like this is to put it into practice and to evaluate it next to your other models,” Shimmin said.
Pricing the Model
One way SpaceX is trying to make Grok 4.5 different from other models is with pricing. SpaceX priced the model at $2 per million input tokens and $6 per million output tokens, claiming it achieves twice the token efficiency of other models. Comparatively, Claude Opus 4.8 is $5 per million input tokens and $25 per million output tokens. OpenAI’s GPT-5.6 Sol, in preview, costs $5 per million input tokens and $30 per million output tokens.
“One of the ways to sustain that competitiveness is by lowering the token cost,” Su said. However, the token cost is unlikely to affect whether enterprises choose Grok 4.5 compared to models from other vendors, Su added. Enterprises that are used to Claude, for example, will most likely not switch to Grok solely because of pricing, he said.
Grok 4.5 is available in Grok Build, in Cursor and from the SpaceXAI console.
