The Grok 4.6 that news sites announced a week early is now real — and it undercuts the models above it by five times
xAI published Grok 4.6 on 12 August, five days after a wave of AI news sites had already written it up as released. We flagged that claim as unconfirmed at the time, because xAI's own model catalogue did not list it. The finished model is aimed at long-running agents and at turning a vague product brief into something that works: xAI says it researches an unfamiliar domain, structures the application, builds the core interactions and keeps refining, testing its own work as tasks get longer. Artificial Analysis puts it at 61 on its Intelligence Index — level with GPT-5.6 Sol, behind Claude Opus 5 at 63 and Fable 5 at 62, ahead of Kimi K3. On agentic tasks its Elo of 1753 trails only Opus 5, and it gets there in about 53 turns and half a billion input tokens where Opus 5 takes roughly 103 turns and two billion. It costs $2 per million input tokens and $6 per million output, against $5 and $25 for Opus 5 and $5 and $30 for GPT-5.6 Sol. It is available in Cursor, Grok Build, the xAI API, OpenRouter, Vercel and Cloudflare. xAI has still not published a parameter count.
Related stories
Rumour: the anonymous model that just topped a coding benchmark, for free, is said to be Zhipu's unreleased flagship2026-08-21DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best
2026-08-21Nvidia is paying $6 billion for the machine that builds a rival's models — and hiring 109 of the people who ran it
2026-08-21Given four hours and a GPU to improve the way AI is trained, the best agent scored 0.25 out of 1 — and most never tried
2026-08-21ChatGPT can now read your iMessages and send them — and the setting that lets it skip asking is the one OpenAI warns about
2026-08-21