Grok 4.5
Grok 4.5 is SpaceXAI’s (xAI) frontier mixture-of-experts (MoE) model engineered specifically for high-efficiency software engineering, long-horizon agentic task execution, and deep technical reasoning. Jointly trained with Cursor using developer interaction traces, Grok 4.5 combines Opus-class coding intelligence with high speed and low output token usage.
Model Overview
Grok 4.5 marks a shift toward specialized developer-agent orchestration, offering high token efficiency and competitive cost-per-task execution.
- Developer: SpaceXAI (xAI)
- Model Tier: Flagship / High-Efficiency Agentic Engine
- Architecture: Mixture-of-Experts (MoE) built on the 1.5-Trillion Parameter V9 Foundation
- Primary Target Use Cases: Multi-file software architecture, terminal/CLI agent execution, autonomous code refactoring, and parallel sub-agent workflows.
Key Capabilities & Performance Profile
Developer-Centric Agentic Training
- Cursor Joint Training: Trained on real-world developer interactions, multi-step debugging traces, and multi-file code editing sessions sourced directly from Cursor.
- High Token Efficiency: Uses up to 4x fewer output tokens than comparable reasoning models on benchmarks like SWE-Bench Pro, drastically reducing task costs during long execution loops.
- Configurable Reasoning Effort: Supports dynamically scalable reasoning effort levels (
low, high, xhigh) to tune latency against logical depth depending on task complexity.
System & Terminal Integration
- Grok Build & CLI Integration: Serves as the primary underlying intelligence for Grok Build and terminal execution environments, optimized for running shell commands and inspecting large repositories.
- Real-time Web & Knowledge Search: Integrates native live search capabilities across web data and social intelligence streams.
Technical Specifications
| Parameter |
Specification |
| Model ID |
grok-4.5 |
| Context Window |
500,000 tokens |
| Input Modalities |
Text, Code, Images (Vision) |
| Output Modalities |
Text, Code, Structured JSON |
| Throughput |
~80–86 tokens/sec |
| Native Capabilities |
Multi-file Code Editing, Parallel Function Calling, Web Search, Terminal/CLI Tools |
Integration Environments
- Cursor IDE: Available as a primary backing model across native Cursor developer plans.
- Grok Build CLI: Used for running autonomous terminal tasks, multi-step code refactoring, and agentic workspace modifications.
- SpaceXAI Console / API: Direct per-token API access via
grok-4.5.
*