- Published on
SpaceXAI Releases Grok 4.5, a Cursor-Trained Model for Coding, Agentic Tasks, and Knowledge Work at $2/M Input
- Authors
- Name
- DeepHorus
SpaceXAI has released Grok 4.5, which it calls its smartest model yet, targeting coding, agentic tasks, and knowledge work. The model was trained alongside Cursor, an AI coding editor, to enhance real engineering capabilities. This release marks a significant step in SpaceXAI's efforts to dominate the AI-assisted development market.
Grok 4.5 is a general-purpose model tuned for software engineering, science, and math. Training involved tens of thousands of NVIDIA GB300 GPUs with advanced stability techniques and data curation including deduplication and quality scoring. The team then scaled reinforcement learning across hundreds of thousands of tasks, focusing on multi-step software engineering.
According to SpaceXAI, Grok 4.5 achieves roughly 4.2 times fewer output tokens than Opus 4.8 on SWE Bench Pro, using 15,954 tokens on average versus 67,020. Pricing is $2 per million input tokens and $6 per million output tokens, served at 80 tokens per second. It ranks first on Harvey's Legal Agent Benchmark, demonstrating office-work strength.
The token efficiency promises lower latency and reduced costs for complex tasks. The model is designed for multi-step reasoning and can handle codebase repair, app prototyping, and spreadsheet modeling. Its availability in Cursor on all plans and via API could accelerate developer adoption.
SpaceXAI's own benchmarks show Fable (max) exceeding Grok 4.5 on all four coding benchmarks, but Grok 4.5 is closest on Terminal Bench 2.1. The company emphasizes its outperformance of comparable leading models in prose. Free usage is offered for a limited time in Grok Build and Cursor.
Grok 4.5 is live in Grok Build and Cursor but not yet in the EU, with expected availability mid-July. It is the default model in Grok Build. As a general-purpose model, it could influence the competitive landscape for coding and knowledge work.
Developers can access Grok 4.5 via the SpaceXAI console with the model ID grok-4.5. The release positions SpaceXAI against rivals like OpenAI and Google in the AI coding assistant space. Continued improvements in token efficiency and benchmark performance may solidify its role in enterprise workflows.
