Grok 4.5: AI for Code at Half the Price of Rivals

What happened?

SpaceXAI has unveiled Grok 4.5, an artificial intelligence model specifically designed for software development tasks and agentic work. The model is available through the SpaceXAI console, Grok Build, and the Cursor editor, owned by Anysphere — whose acquisition by SpaceX was announced in June 2025 for $60 billion. Grok 4.5 was co-trained with Cursor using trillions of tokens of user interaction data with codebases and software tools. According to Cursor, the model was co-trained with SpaceXAI using data from user interactions with codebases and software tools, giving it deep knowledge of real-world coding patterns.

The pricing is notably low: $2 per million input tokens and $6 per million output tokens, with a speed of 80 tokens per second. According to Artificial Analysis's Coding Agent Index, Grok 4.5 completes coding tasks at an average cost of $2.49 per task, compared to $5.07 for GPT-5.5 on Codex and $11.80 for Fable 5 on Claude Code. In accuracy, it sits slightly below Fable 5 but roughly on par with GPT-5.5. This represents a cost reduction of up to 79% compared to Fable 5 and 51% compared to GPT-5.5.

The launch comes amid a context where the costs of AI agents for coding have rapidly escalated. A June 2025 InfoWorld report already warned that AI coding token costs were on track to rival human payrolls. For example, running an autonomous coding agent for a full workday could generate token bills exceeding $500 per day, according to analyst estimates. Grok 4.5 aims to reverse that trend.

Why is it important?

The launch addresses a growing problem in enterprises: the cost of AI agents for coding has skyrocketed, rivaling human salaries. Neil Shah, vice president at Counterpoint Research, noted that "companies are hitting a wall with AI ROI" due to "massive token bills." Grok 4.5 offers immediate relief by cutting the cost per task to half or less than its direct competitors. Shah added that "AI adoption is becoming a one-way, expensive street" and that models like Grok 4.5 are necessary for sustainable ROI.

Moreover, the integration into Cursor gives SpaceXAI a direct channel to developers already using that environment, rather than relying solely on an API. This could accelerate enterprise adoption, though analysts like Anand Joshi of JP Data warn that "it's too early to say if Grok 4.5 is a game-changer" and that the developer community will deliver the final verdict. Joshi also noted that "cost per task is the metric that matters, not cost per token," and that companies should test the model on their own workflows.

The historical context is relevant: in 2024, the launch of models like Claude 3.5 Sonnet and GPT-4o boosted developer productivity but also raised concerns about recurring costs. Grok 4.5 is the first model explicitly designed to minimize token spending, which could mark a turning point in the economics of AI applied to software.

Market implications

• Competitive pressure on OpenAI and Anthropic: Grok 4.5's pricing will force rivals to justify their higher costs or reduce them. OpenAI has already responded with volume discounts on GPT-5.5, but the gap remains significant. Anthropic, for its part, has announced efficiency improvements in Fable 5 but has not changed its base prices.
• Accelerated enterprise adoption: The lower cost per task could drive AI usage in development teams that previously considered it too expensive. According to a May 2025 Gartner survey, 62% of companies cited cost as the main barrier to adopting AI agents for coding. Grok 4.5 could remove that obstacle.
• Ecosystem consolidation: SpaceX's purchase of Anysphere and the integration of Grok into Cursor reinforce the trend of big tech controlling the development toolchain. This echoes Microsoft's acquisition of GitHub in 2018, which integrated Copilot into developers' workflows. However, the difference is that SpaceX now controls both the model and the editor, which could create technological dependency.
• Impact on AI startups: Smaller companies competing in the AI coding space, such as Codeium or Replit, may face pressure to cut prices or find niches. Replit has already announced an integration with Grok 4.5 on its platform, according to internal sources.

What readers should know

Cost per token is not everything: cost per successful task is the key metric. Companies should test Grok 4.5 on their own codebases before migrating. Availability in the EU is expected by mid-July. For now, the model is available on the SpaceXAI console, Grok Build, and Cursor. Additionally, SpaceXAI has announced that the model will be available on Azure AI Foundry and Google Cloud Vertex AI in the coming weeks, expanding its enterprise reach.

Developers should note that while Grok 4.5 is cheaper, its accuracy on complex tasks is slightly lower than Fable 5. For critical tasks, combining models may be necessary. It's also important to consider that the model was trained on Cursor data, which could give it an advantage in similar environments but perhaps not in other editors.

Regarding privacy, SpaceXAI has assured that user data is not used to train models without consent, though the integration with Cursor means Anysphere has access to coding data. Companies should review data policies before adopting the solution.

Finally, the launch of Grok 4.5 is a milestone in the race to make AI economically viable for software development. If other players respond with similar price cuts, we could see a democratization of access to AI agents, but also pressure on AI companies' margins. Time will tell whether SpaceXAI's strategy of prioritizing volume at low cost is sustainable in the long term.