TradingKey - SpaceXAI (SPCX) officially launched Grok 4.6, further doubling down on long-running agents, complex programming, and knowledge work scenarios. Upgraded from Grok 4.5, the model focuses on enhancing sustained execution capabilities for multi-step tasks, suitable for thematic research, information analysis, cross-repository collaboration, and converting product ideas into runnable applications or deliverables.

SpaceXAI stated that Grok 4.6 underwent a longer period of supplementary training, with training data covering reasoning, advanced technical concepts, and high-quality engineering content. Subsequent training also covered various agent reinforcement learning tasks such as knowledge work, general programming, kernel optimization, web development, and computer-aided design.

The company stated that the new model demonstrates stronger self-testing and verification capabilities in long-running tasks, and can also complete initial product frameworks and visual designs faster in interactive and visual projects.

According to benchmark results released by the company, Grok 4.6 High achieved a overall score of 61 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Sol Max and higher than Grok 4.5 High's 56 points. It also achieved top scores on the benchmark table in evaluations including GDPVal-AA v2, AA-Briefcase, and Harvey LAB.

Source: SpaceXAI

Grok 4.6 is currently integrated into Cursor and Grok Build, offering double usage quotas during its first week; it is also available via platforms such as the API, OpenRouter, Vercel, and Cloudflare. The standard version is priced at $2 per million input tokens and $6 per million output tokens, with a high-speed version available at twice the price of the standard version.

Find out more