SpaceXAI launches Grok 4.6, a 500K‑context workhorse for long‑running AI agents

SpaceXAI, Elon Musk’s AI division inside SpaceX, has rolled out a new large language model called Grok 4.6. The model is an evolutionary update to Grok 4.5 and is tuned for long‑running AI agents that handle complex coding and knowledge‑work tasks. A key feature is its 500,000‑token context window, large enough to span entire codebases or multi‑day agent sessions without constantly truncating history.

According to materials shared with developers, Grok 4.6 accepts both text and images, supports web and X (旧Twitter) search, tool and function calling, and code execution. Independent benchmarking discussions suggest it is edging close to the performance of OpenAI’s GPT‑5.6 Sol tier while undercutting many competitors on price, with a base rate around $2 per million input tokens and $6 per million output tokens, and higher pricing only for very long contexts.

Developer reaction is mixed. Many praise Grok 4.6 as a “great model” for fast, end‑to‑end automation, but some criticize the 500K context as insufficient for the most demanding architectural tasks, calling for 1M tokens or more. Others remain wary of Grok’s safety profile and political leanings, noting that more capable agentic models increase the urgency of robust governance and misuse safeguards. With SpaceXAI signaling further context and efficiency upgrades ahead, Grok 4.6 strengthens its position as a serious alternative in the high‑end agent and coding market.