Renamed SpaceXAI Launches Grok 4.5 Coding Model
Elon Musk's newly restructured artificial intelligence unit, SpaceXAI, released its latest large language model, Grok 4.5. The company built this mixture-of-experts software system specifically to handle software engineering, data science, legal work, and financial tasks.
The release represents the first major model launch since Musk folded his AI company, xAI, into SpaceX earlier this year. The launch also follows SpaceX's recent 60 billion dollar acquisition of the AI coding platform Cursor in mid-June.
The corporate logo displays the model name beneath the stylized xAI symbol. This identifier highlights the legacy design of the system before the recent corporate integration into SpaceX.
SpaceXAI trained the model jointly with Cursor, utilizing trillions of tokens of developer interaction data. This training dataset records how software engineers write code and how automated agents interact with code repositories.
Musk publicized the system release through a post on the social media network X.
According to Musk's post on X:
"It is an Opus-class model, but faster, more token-efficient and lower cost,"
The new system runs on a 1.5-trillion-parameter foundation architecture called V9. The company states that the V9 base is three times larger than their previous system architecture.
The model currently has a context window of 500,000 tokens and processes data at an approximate speed of 80 tokens per second. Musk indicated that the current speed does not represent the technical limit for the system.
According to a subsequent statement by Musk on X:
"Grok 4.5 is not yet using our internally developed C/C++ inference software that exact maps to the GB300 hardware. Doubling or more of the current speed is probably achievable."
Benchmark results from Datacurve show varied performance across industry evaluations. On the DeepSWE 1.0 test, Grok 4.5 scored 62.0 percent, trailing Fable at 66.1 percent and GPT 5.5 at 64.31 percent.
On the SWE Marathon resolution rate benchmark, Grok 4.5 recorded a score of 29.0 percent. This score places it ahead of Claude Opus 4.8 at 26.0 percent and Fable at 24.0 percent.
Cursor disclosed in a corporate blog post that an early snapshot of its own codebase was accidentally included in the training data mix. The company stated that the exact impact on its internal CursorBench benchmark remains unclear.
The company priced the base model at 2 dollars per million input tokens and 6 dollars per million output tokens. A faster variant costs 4 dollars per million input tokens and 18 dollars per million output tokens.
The launch occurred on the same day that competitor OpenAI launched its new conversational voice software, ChatGPT 5.6. OpenAI priced its new model at 1 dollar per million input tokens and 6 dollars per million output tokens.
Anthropic charges 5 dollars per million input tokens and 25 dollars per million output tokens for its Claude Opus 4.8 model. SpaceXAI internal testing states that Grok 4.5 performs roughly at the level of Claude Opus 4.7 but at lower operational expenses.
According to a report by the Washington Examiner, Musk stated:
"Our internal assessment is that Grok 4.5 is roughly comparable to Opus 4.7, but much faster,"
He added:
"The combination of capability, faster speed and lower cost is what makes it competitive."
The model is available immediately through the SpaceXAI API console, the Grok Build platform, and all Cursor subscription plans. The system is not accessible to users inside the European Union.
SpaceXAI plans to expand availability to European markets later this month. The company has announced a product development schedule that includes monthly model releases through the remainder of 2026.
