Full story
Grok 4.5 Launch Details
SpaceXAI released Grok 4.5, calling it its “smartest model to date” and positioning it for coding, agentic tasks, and knowledge work.
SpaceXAI says Grok 4.5 was trained alongside Cursor and that training ran across “tens of thousands of NVIDIA GB300 GPUs.”
The company priced Grok 4.5 at $2/M input and $6/M output, and said it is served at 80 TPS.
In its own launch materials, SpaceXAI also said Grok 4.5 is now the default model in Grok Build and is available in Cursor on all plans and from the SpaceXAI console.
SpaceXAI added that “It is not yet available in the EU,” with EU availability expected in mid-July.
Musk Frames Opus-Class
Elon Musk said Grok 4.5 is “an Opus-class model, but faster, more token-efficient and lower cost,” after SpaceXAI reported “strong positive feedback” from its beta test program.
SiliconANGLE reported that SpaceXAI claims Grok 4.5 can do tasks “just as well as its peers” at around half the cost, citing “twice greater token efficiency.”
Musk also compared Grok 4.5 to Anthropic’s Opus line on X, and SiliconANGLE said his follow-up described internal assessments as “roughly comparable” with Opus 4.7 in performance but “much faster.”
For pricing context, SiliconANGLE put Opus 4.7 at $5 per one million input tokens and $25 per one million outputs, while SpaceXAI’s Grok 4.5 was described as $2 per one million input tokens and $6 per one million output tokens.
The launch materials from SpaceXAI said Grok 4.5 “delivers intelligent results to you more quickly and at far lower costs” by combining “80 TPS” service with “twice greater token efficiency.”
Benchmarks, Use Cases, and Next
SpaceXAI’s launch materials said Grok 4.5 “exceeds comparable leading models” at real engineering tasks and described its reasoning as “intelligent and efficient.”
“SpaceXAI has released Grok 4”
The company said it trained Grok 4.5 on datasets spanning “knowledge in coding, science, engineering, and math,” and that reinforcement learning covered “hundreds of thousands of tasks” centered on multi-step software engineering.
SpaceXAI also tied the model’s agentic rollout approach to its training stack, stating “agentic rollouts can run for many hours while learning continues across tens of thousands of GPUs.”
For concrete applications, SpaceXAI’s examples included building a Three.js solar-system simulation from one prompt and drafting a “5-slide quarterly business review Q3 Business Review Revenue, margins, pipeline — and where we invest next” in PowerPoint.
Looking ahead, SpaceXAI said it is offering “free Grok 4.5 usage for a limited time in Grok Build and Cursor,” while also stating that EU availability is expected in mid-July and that the model is not yet available in the EU.
