← All News

xAI Ships Grok 4.6 the Same Night: Frontier Scores at $2 / $6

Source: xAI

xAI released Grok 4.6 on 12 August 2026, hours after DeepSeek's V4 Pro GA build. xAI's docs now describe it as the most intelligent and fastest model the company has built, aimed explicitly at long-running agents and interactive, visual work rather than single-shot chat.

Where It Lands

On xAI's published comparisons, Grok 4.6 scores 61 on the Artificial Analysis Intelligence Index — level with GPT-5.6 Sol, and up from 56 for Grok 4.5. It leads on CursorBench (69.9), DeepSWE (65.9) and APEX-Agents (57.5), while trailing on Terminal-Bench and several other evaluations. xAI highlights improved self-testing and verification over long task sequences, which is the failure mode that usually ends multi-hour agent runs.

The Price Is the Story

Grok 4.6 is $2 per million input tokens and $6 per million output, with a 500K context window; both rates double once a prompt passes 200K tokens, and a faster variant costs twice the base rate. That places frontier-adjacent scores several times below the flagship tier from OpenAI and Anthropic. It is available through the xAI API, Grok Build, Cursor, and partners including OpenRouter, Vercel and Cloudflare; xAI is running a 2x usage bonus in Grok Build and Cursor for the first week.

Two Releases, One Squeeze

Taken together, the two launches compress the middle of the market from both ends on the same night. DeepSeek V4 Pro puts near-frontier agent scores at $0.87 per million output tokens if you can accept slower runs and text-only input. Grok 4.6 asks $6 and gives back speed, vision and a subscription route. A model that is neither the cheapest nor the fastest nor the strongest now has a much harder case to make — and the practical takeaway for anyone running agents is that the right answer is rarely one model. Cheap asynchronous work and interactive work have different winners this week, and both were different a month ago.

Both models are in the picker on OpenClaw Launch, and you can switch between them from the dashboard or with the /model command without redeploying. Setup details are in the Grok guide and the DeepSeek V4 Pro guide.

Build with OpenClaw

Deploy your own AI agent in under 30 seconds — no servers, no CLI.

Deploy Now