SpaceXAI unveiled Grok 4.5, a new mixture‑of‑experts LLM trained on trillions of real‑world coding tokens. It targets complex software‑engineering, autonomous‑agent workflows and broader enterprise knowledge tasks, priced at $2 / M input and $6 / M output tokens.
SpaceXAI today unveiled Grok 4.5, its most capable large‑language model yet, engineered for deep software‑engineering work, autonomous agent workflows, and enterprise‑level knowledge tasks.
A new architecture for professional AI
Grok 4.5 runs on a mixture‑of‑experts (MoE) backbone and was trained on trillions of tokens harvested from real‑world coding sessions and developer pipelines via Cursor. Instead of merely ingesting static source code, the model learned from actual engineering patterns, giving it a nuanced grasp of how developers interact with tools, repositories, and AI assistants.
Designed for complex, multi‑step tasks
The model excels at long‑running, multi‑step engineering problems that demand sustained reasoning. Beyond code, SpaceXAI markets Grok 4.5 for data‑science pipelines, financial modeling, legal research, cybersecurity analysis, and other enterprise knowledge workloads.
Pricing aimed at scale
To make high‑performance AI affordable for businesses, SpaceXAI set the price at $2 per million input tokens and $6 per million output tokens, positioning Grok 4.5 as a cost‑effective alternative for large‑scale deployments.
The release arrives amid a heated race among AI vendors to boost coding assistance, autonomous agents, and large‑scale reasoning capabilities. Grok 4.5 signals SpaceXAI’s shift from pure conversational AI toward robust professional automation.
Comments
No comments yet.