SpaceXAI launched Grok 4.5, its latest artificial intelligence model designed for coding, agentic workflows and advanced knowledge work. The company positions Grok 4.5 as its most capable system to date, built for technical tasks including software engineering, data science, research, finance and multi-step professional work. The launch represents Elon Musk's effort to move Grok beyond consumer chatbot functionality into the enterprise AI market, where model providers compete on performance, speed, token efficiency and cost. The AI industry is shifting focus toward coding agents and workplace automation, with enterprise customers prioritizing inference cost, latency and reliability as they deploy AI tools across engineering, customer support, research, legal and financial workflows.
SpaceXAI Prices Grok 4.5 Below Anthropic's Claude Opus
SpaceXAI priced Grok 4.5 at $2 per million input tokens and $6 per million output tokens. This undercuts Anthropic's Claude Opus 4.8 pricing of $5 per million input tokens and $25 per million output tokens. Musk described Grok 4.5 as "Opus-class" but faster, more token-efficient and cheaper in a direct comparison with Anthropic's Claude Opus family. The company said Grok 4.5 is served at fast-model speeds of about 80 tokens per second and uses fewer output tokens on certain software-engineering tasks than competing high-end models. Lower token usage can reduce costs for companies running large volumes of AI queries, particularly in coding and agentic workflows where models may generate long outputs or perform many tool calls.
Grok 4.5 Trained on Nvidia GB300 GPUs with Cursor Integration
SpaceXAI said Grok 4.5 was trained across tens of thousands of Nvidia GB300 GPUs and used data filtering, deduplication and quality-scoring techniques aimed at improving training efficiency. The company said the model was trained alongside Cursor, the AI coding platform. Grok 4.5 is available through Grok Build, Cursor and the SpaceXAI developer console. The model's availability through Cursor is strategically important because coding environments have become one of the most valuable distribution channels for AI models, as developers use them directly inside daily workflows. The company's benchmark disclosures highlighted performance across software-engineering evaluations, including DeepSWE, SWE Marathon, Terminal Bench and SWE Bench Pro.
SpaceXAI Competes on Cost and Speed in Enterprise AI Market
The Grok 4.5 launch shows how the AI race is shifting toward coding agents and workplace automation rather than general chat alone. SpaceXAI says the model is designed to handle long-duration, multi-step tasks, including engineering projects that require planning, code generation, debugging and iterative problem-solving. The release strengthens Musk's broader AI strategy after xAI was folded into SpaceX and rebranded under the SpaceXAI banner. The company is now competing more directly with OpenAI, Anthropic and Google, while also leveraging large-scale compute infrastructure tied to Musk's broader technology ecosystem. OpenAI, Anthropic and Google are all pushing more advanced systems for enterprise and developer use, while smaller providers are trying to compete on specialization and price.
FAQ
What is the pricing for SpaceXAI's Grok 4.5 AI model?
SpaceXAI priced Grok 4.5 at $2 per million input tokens and $6 per million output tokens, which undercuts Anthropic's Claude Opus 4.8 pricing of $5 per million input tokens and $25 per million output tokens.
How was Grok 4.5 trained and where is it available?
Grok 4.5 was trained across tens of thousands of Nvidia GB300 GPUs using data filtering, deduplication and quality-scoring techniques. The model was trained alongside Cursor and is available through Grok Build, Cursor and the SpaceXAI developer console.
What tasks is Grok 4.5 designed to handle?
Grok 4.5 is designed for technical tasks including software engineering, data science, research, finance and other multi-step professional work. The model handles long-duration tasks including engineering projects that require planning, code generation, debugging and iterative problem-solving.
