SpaceXAI introduces Grok 4.5, emphasizing cost and speed, while aiming to compete effectively in the AI market.
SpaceXAI has announced the release of Grok 4.5, positioning it as the company's most advanced model to support programming and agent-related tasks. Emphasizing cost-effectiveness and speed, Grok 4.5 is set to compete more aggressively in the AI landscape.
Grok 4.5 marks our first model tailored specifically for coding, enhanced with Cursor training, delivering unprecedented intelligence at high speeds and competitive pricing.https://t.co/i8HpU7w64k pic.twitter.com/oBjGtTsoNc
— SpaceXAI (@SpaceXAI) July 8, 2026
A variety of access points, including API and Grok Build, are available for users at different pricing levels. However, it is important to note that Grok 4.5 will initially not be available in the EU, with plans for its launch set for mid-July.
The pricing structure underscores Grok 4.5's competitive edge, charging $2 per million input tokens and $6 per million output tokens. In contrast, comparable models from other vendors, such as Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.6 Sol, are priced significantly higher.
According to Elon Musk, Grok 4.5 is an "Opus-class model, yet more efficient and accessible," suggesting that SpaceXAI's internal evaluations indicate it performs similarly to Opus 4.7, but with greater speed.
Following positive feedback from our beta users, we are thrilled to announce Grok 4.5's public release tomorrow, offering a blend of speed, efficiency, and affordability.
— Elon Musk (@elonmusk) July 8, 2026
Performance Metrics
Initial testing results for Grok 4.5 indicate a varied performance landscape. In the DeepSWE 1.0 test, Grok achieved a score of 62%, trailing behind some competitors but outperforming others. Subsequent tests showed similar trends with Grok maintaining a competitive stance overall.
In deeper evaluations, Grok 4.5 scored 53% in DeepSWE 1.1 and performed strongly in the SWE Marathon with a leading score of 29% against key competitors. Additionally, Grok excelled in Harvey's Legal Agent Benchmark.
Cost Efficiency Argument
SpaceXAI highlights that Grok 4.5's economic advantages are noteworthy. In SWE Bench Pro tasks, Grok utilized significantly fewer output tokens than Claude Opus 4.8, suggesting potential cost savings for users engaged in extensive task cycles.
With an output rate of approximately 80 tokens per second and a broad context window of 500,000 tokens, Grok 4.5 aims to strike an optimal balance between performance, cost, and generation speed.
Training Insights
The model was developed in conjunction with Cursor, enhancing its focus on programming, engineering, and mathematics tasks. While the complete training data has not been disclosed, it included practical data from developer sessions rather than static code.
In June, SpaceXAI finalized a substantial acquisition deal for Cursor, valued at $60 billion, transitioning into regulatory approval phases. Grok 4.5 represents a significant milestone following Musk's integration of the AI division with SpaceX, leveraging substantial GPU resources for its development.
Earlier this year, plans emerged regarding a collaboration between SpaceX and xAI to produce software aimed at the Pentagon's autonomous systems, indicating broader ambitions in military applications.