SpaceXAI unveils Grok 4.6: How does the new AI model compare to competitors?

SpaceXAI, Elon Musk's AI startup, has released Grok 4.6, a new AI model that builds on its predecessor Grok 4.5 but emphasizes longer-span agentic tasks. In the Artificial Analysis Intelligence Index, the model scored 61 points, matching OpenAI's GPT-5.6 Sol, while Anthropic's Claude Fable and Claude Opus scored 62 and 63 points respectively. Grok 4.6 performed strongly on the 'GDPval-AA v2' benchmark with an Elo score of 1753, just behind Claude Opus 5. The model is trained on a variety of agentic tasks including knowledge work, coding, and computer-aided design, and is particularly adept at turning a product idea into a working first version, researching unknown areas, structuring applications, and refining results through feedback loops. Pricing is a key advantage: Grok 4.6 costs $2 per million input tokens and $6 per million output tokens, over 60% cheaper than Claude Opus 5 ($5 and $25) and significantly cheaper than GPT-5.6 Sol ($5 and $30). Musk claimed on X that Grok 4.6 is 'objectively number one' considering intelligence, speed, and cost. A day prior, SpaceXAI released the Grok Bot, a digital colleague that runs tasks autonomously in the background and only returns when needing approval or finishing a job.