Introducing Grok 4.6

Summary

Grok 4.6, released with SpaceXAI, is presented as a model for long-running agents and ambitious interactive, visual, coding, and knowledge-work tasks. It builds on Grok 4.5 through a longer supplemental training run using curated model-generated reasoning and technical data, an improved optimizer and training recipe, regenerated SFT trajectories, model-based filtering, and agentic reinforcement-learning tasks across several environments. The release says Grok 4.6 matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index and reports stronger first passes, sustained multi-step work, self-testing, and iterative refinement for working projects. It is available in Cursor, Grok Build, the SpaceXAI API, and through OpenRouter, Vercel, and Cloudflare, with pricing from $2 per million input tokens and $6 per million output tokens, while a fast variant costs twice as much.

Takeaways

1. 1

   Grok 4.6 matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index, a composite score based on nine benchmarks.

2. 2

   The model is reported to produce stronger first passes on visual and interactive projects, while also showing more self-testing and verification on longer trajectories.

3. 3

   Availability includes Cursor, Grok Build, the SpaceXAI API, OpenRouter, Vercel, and Cloudflare; pricing starts at $2 per million input tokens and $6 per million output tokens.