xAI shipped Grok 4.6, its latest foundation model designed for sustained agentic work across complex, multi-step tasks.

The model matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index with a score of 61, tying for second place behind Fable 5's 62. Grok 4.6 shows particular strength in coding benchmarks, scoring 65.9% on DeepSWE v1.1 and 69.9% on CursorBench v3.2.

Grok 4.6 underwent extended supplemental training with curated model-generated data for reasoning and technical concepts. The company used Grok 4.5 to regenerate supervised fine-tuning trajectories across reasoning, agent frameworks, and domains including STEM and software engineering.

The model excels at transforming broad product ideas into working prototypes. Internal testing showed Grok 4.6 can research unfamiliar domains, structure applications, implement core interactions, and refine results through multiple feedback rounds.

Stronger visual and interactive capabilities

Grok 4.6 produces more polished first attempts on visual and interactive projects compared to its predecessor. The model establishes application structure and visual language in single passes, making it particularly useful for rapid prototyping workflows.

The model demonstrates improved self-verification behaviors on longer tasks, checking its own work before proceeding to subsequent steps.

Grok 4.6 launches today in Cursor and Grok Build with 2x included usage for the first week. API pricing starts at $2 per million input tokens and $6 per million output tokens, with a fast variant available at double the price.

The model is also available through OpenRouter and other integration partners including Vercel and Cloudflare. xAI conducted its most extensive pre-deployment safety testing suite to date, calibrating safeguards for the model's expanded capabilities across legitimate use cases.