SpaceXAI releases Grok 4.5, offering an “Opus-class” model for coding and office tasks

When SpaceXAI shipped Grok 4.5 this week, the pitch was not that it beats the best models. It was that it does comparable work while…Continue reading on Towards AI »

SpaceXAI has launched its most intelligent model, Grok 4.5, for complex tasks. This advanced AI model excels in software engineering and data science applications. Grok 4.5 is an Opus-class model that is faster and more efficient. It is available today across multiple platforms with specific pricing structures. EU availability is anticipated later this month for the new AI model.

SpaceXAI, the company behind the Grok chatbot, has launched Grok 4.5. The company describes it as its smartest model yet, built for coding, agentic tasks and knowledge work. According to the official announcement, Grok 4.5 was trained alongside Cursor, the AI coding tool.Built for engineering and office tasksGrok 4.5 was trained on datasets spanning coding, science, engineering and mathematics. SpaceXAI says the model combines intelligent and efficient reasoning to excel at real world engineering tasks, outperforming comparable leading models on several benchmarks. The company also says Grok 4.5 ranks first on Harvey's Legal Agent Benchmark, indicating strong performance on office and knowledge work beyond coding.SpaceX Acquires AI Coding Startup Cursor In $60 Billion All-Stock Deal To Compete With OpenAI And AnthropicKey features- Default model in Grok Build, where it can construct complex Excel models involving web research and multi sheet formula use- In PowerPoint and Word, the model can use native PowerPoint shapes to build diagrams, design slide content and write clear prose- Served at 80 tokens per second, which SpaceXAI describes as fast model speed- Twice the token efficiency of the latest leading models on the same tasks, according to the company- Resolves tasks using an average of 15,954 output tokens on SWE Bench Pro, about 4.2 times fewer tokens than Opus 4.8 in max mode, per SpaceXAI's own benchmark data- Trained across tens of thousands of NVIDIA GB300 GPUs- Priced at $2 per million input tokens and $6 per million output tokensElon Musk's xAI Rebrands To SpaceXAI, Unveils New Logo Training approachSpaceXAI says it invested heavily in data filtering and curation, including deduplication, quality scoring and domain focused selection, to keep the training data mixture high coverage and high signal. The company also says it scaled reinforcement learning with a focus on per token intelligence, with training covering hundreds of thousands of tasks centred on multi step software engineering and other technical work.Grok 4.5 availabilityGrok 4.5 is available starting today in Grok Build, in Cursor across all plans, and through the SpaceXAI API console. Developers can access the model using an API key. SpaceXAI has noted that Grok 4.5 is not yet available in the European Union in any of its products or through the API console, with EU availability expected in mid July.

SpaceXAI's new coding model is cheaper and faster than Anthropic and OpenAI's flagships—and by Musk's own account, at least one generation behind them.

Elon Musk's tech company released the newest version of Grok on Wednesday, promising a cheaper, more efficient alternative to other powerful AI models.

Elon Musk said the latest version of Grok is “faster, more token-efficient and lower cost” than one of Anthropic’s models.