Grok 4.5 Debuts as a Tamer, Office-Ready AI Model

SpaceXAI, the company formerly known as Eloncorp, released Grok 4.5 on Wednesday, positioning the latest version of its AI model as a serious tool for coding, legal work, and office productivity. The company claims the model is cheaper and more efficient than competing offerings.

According to SpaceXAI, Grok 4.5 is “the smartest model built to excel at coding, agentic tasks, and knowledge work.” The model scored number one on Harvey’s Legal Agent Benchmark, a measure of legal reasoning, and is designed to handle tasks such as building complex Excel models that involve web research, multi-sheet formulas, and leaving notes for future reference.

The release marks a sharp pivot from Grok’s controversial past. Earlier versions of the model were known for generating inappropriate content, including deepfake porn and a persona called “MechaHitler,” which led to temporary bans in some regions.

The model was trained on “tens of thousands” of Nvidia GB300 GPUs, with a focus on data quality over quantity. The company invested heavily in deduplication, quality scoring, and domain-specific selection. Reinforcement learning, a technique also used by OpenAI and DeepSeek, was employed to teach the model hundreds of thousands of tasks, reducing the number of thinking tokens needed for complex problems. SpaceXAI says this results in faster serving speeds of up to 80 tokens per second.

Independent benchmarks from Artificial Analysis show that Grok 4.5 roughly matches OpenAI’s GPT-5.5 and Claude Opus 4.8, though it still trails Anthropic’s Claude Fable. Pricing is aggressive: $2 per million input tokens and $6 per million output tokens, compared to GPT-5.5’s $5/M input and $30/M output.

Grok 4.5 is available immediately in Grok Build, Cursor, and the SpaceXAI console for users outside the European Union. EU rollout is expected in mid-July.