Grok 4.6 expands to Vertex AI, Bedrock and Cursor, tops CursorBench

Grok 4.6, xAI's successor to Grok 4.5, is executing an aggressive distribution land-grab. It arrived on Google Cloud Vertex AI with a 500K-token context window and configurable reasoning efforts, following its August 19 addition to Amazon Bedrock with cross-region inference routing across US and global regions, and is also live on Cursor, OpenRouter, Vercel and Cloudflare. On Bedrock it's billed as a 500K-context coding and agentic model.

The headline benchmark claim is #1 on CursorBench in 'Extra High Thinking' mode — a result developers greeted with both enthusiasm and scrutiny over benchmark timing. Pricing is the other draw: at roughly $2/$6 per 1M input/output tokens (~$8 combined), Grok 4.6 undercuts Opus and Sonnet's $30–60 range, positioning it as a cost disruptor for agent workloads. The friction point flagged by developers is a long-context billing cliff that doubles at 200K tokens.

xAI paired the model rollout with wider access to Grok Bot, a persistent cloud agent service (reported at $300/month) now offering a seven-day free trial. The combination — cheap frontier-ish model plus productized autonomous agent — is xAI's bid to win developer mindshare across every major cloud and IDE simultaneously rather than through a single flagship channel.

The strategic read is that distribution breadth, not raw benchmark leadership, is xAI's real weapon here: by shipping to Vertex, Bedrock, Cursor, OpenRouter, Vercel and Cloudflare at once, it maximizes the chance of default adoption before enterprises complete evaluations. The open questions are benchmark durability under independent testing and whether the 200K billing cliff undermines the cost story for exactly the long-horizon agent tasks the 500K window is meant to serve.