SpaceXAI released Grok 4.5 on July 8, pricing it at $2 per million input tokens and $6 per million output tokens — undercutting Anthropic's Claude Opus 4.8 ($5/$25) and landing close to OpenAI's budget-tier GPT-5.6 Luna ($1/$6), according to the company's own announcement and coverage from Forbes and Axios. SpaceXAI is the rebranded entity that resulted from SpaceX's February merger with xAI, a share-exchange deal that combined the two companies at a reported $1.25 trillion valuation.

A Cursor-trained model

Grok 4.5 is the first model SpaceXAI has trained using data from Cursor, the AI coding tool that xAI's parent is in the process of acquiring for roughly $60 billion. Cursor published its own companion blog post confirming the model is available inside its editor alongside the xAI API. SpaceXAI and outlets including InfoWorld and MarkTechPost describe the model as built specifically for coding, agentic workflows, and knowledge-work tasks like spreadsheets and documents, rather than as a general-purpose conversational chatbot.

Efficiency over raw pricing

Beyond the sticker price, coverage has focused on token efficiency: benchmark write-ups citing SWE-Bench Pro results reported Grok 4.5 using an average of roughly 16,000 output tokens per task, versus roughly 67,000 for Claude Opus 4.8 on the same benchmark — meaning the effective cost gap between the models is larger than the per-token price difference alone suggests. Elon Musk has said the model competes with last year's Claude Opus generation on capability. As with any vendor-supplied benchmark, these figures haven't been independently reproduced by a third party.

Why it matters

The launch lands one day after OpenAI took GPT-5.6 fully public and in the same week Anthropic's Claude Sonnet 5 became the default model for all users — making early-to-mid July a dense stretch of frontier-model releases. Grok 4.5's pitch is less about topping capability leaderboards and more about cost-per-completed-task in coding and agentic use cases, a framing that puts pressure on rivals to justify higher per-token prices with fewer retries and shorter task loops rather than raw benchmark scores alone.