Grok 4.6
SpaceXAI's frontier model for coding, agentic tasks, and knowledge work, focused on long-running agents and ambitious interactive and visual projects
Grok 4.6
SpaceXAI • August 2026
Training Data
Curated model-generated reasoning data and high-quality engineering data
Grok 4.6
August 2026
Parameters
1.5 trillion (reported)
Training Method
Longer supplemental training than Grok 4.5, plus agentic RL across knowledge work, coding, kernel optimization, web development, and CAD
Context Window
500,000 tokens
Knowledge Cutoff
Not disclosed
Key Features
500K Token Context • Self-Testing & Verification • Visual & Interactive Work • Reasoning Effort: Low to xhigh
Capabilities
Coding: Outstanding (65.9% DeepSWE v1.1)
Knowledge Work: Outstanding (1753 GDPval-AA v2)
Agentic: Outstanding (69.9% CursorBench v3.2)
What's New in This Version
Matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index (61) at Grok 4.5's price ($2/$6 per 1M tokens), with stronger self-testing and better at turning product ideas into working applications
SpaceXAI's frontier model for coding, agentic tasks, and knowledge work, focused on long-running agents and ambitious interactive and visual projects
What's New in This Version
Matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index (61) at Grok 4.5's price ($2/$6 per 1M tokens), with stronger self-testing and better at turning product ideas into working applications
Technical Specifications
Key Features
Capabilities
Other SpaceXAI Models
Explore more models from SpaceXAI
Grok 4.7
SpaceXAI's best model to date — a larger Grok trained on a longer, harder reinforcement-learning run plus SpaceX engineering data, at Grok 4.6's price and speed
Grok 4.5
SpaceXAI's coding and agentic-work flagship on the new V9 foundation — pitched as Opus-class capability but faster, more token-efficient, and lower cost
Grok 4.3
xAI's reasoning flagship with always-on reasoning, native video input, 1M context, and significantly cheaper API pricing
