Grok 4.6: SpaceXAI ships agent model with 500K context

Elon Musk's AI unit — referred to as SpaceXAI in current coverage after xAI moved under the SpaceX umbrella — has released Grok 4.6. The model ships with a 500,000-token context window, accepts text and image input, and is explicitly tuned for long-running agent workflows, coding and knowledge-intensive work.

On benchmarks, Grok 4.6 lands near the top: it scores 61 on the Artificial Analysis Intelligence Index, matching OpenAI's GPT-5.6 Sol — enough, according to VentureBeat, for third place worldwide, ahead of Kimi K3. On the coding tests that matter most to engineering teams, a gap remains: Grok 4.6 reaches 65.9 percent on DeepSWE v1.1, versus 73 percent for GPT-5.6 Sol.

Technically this is not a bigger base model but a post-training upgrade: SpaceXAI leaned on a longer supplemental training run and reinforcement learning in agentic environments. That is where the claimed strengths come from — repository refactors, migration agents, research synthesis over large document sets, and finance and legal analysis.

The pricing is the aggressive part: $2 per million input tokens, $0.50 for cached input and $6 per million output tokens — for prompts under 200,000 tokens, with rates doubling above that threshold. The model is available via the xAI API, Cursor and Grok Build, and can be routed through OpenRouter, Vercel and Cloudflare. There are no open weights and no self-hosting option.

For European teams, Grok 4.6 mostly means one thing: fresh price pressure in the agent segment. Anyone running agent pipelines on pricier frontier models now has a serious alternative — but has to weigh the weaker coding scores and the missing self-hosting option against the cost advantage.