What changed
SpaceXAI released Grok 4.6 in August, only weeks on from Grok 4.5, which means frontier models now ship monthly from Elon Musk's AI arm. The launch report shows a model tuned to stay on task through long stretches of work: code, research, office jobs with less hand-holding.
On the Artificial Analysis Intelligence Index, the data shows the new model scoring 61 in August, up five points from Grok 4.5, which means it ties GPT-5.6 Sol Max for third behind Anthropic's two leaders.
Grok 4.6's August score on the Artificial Analysis Intelligence Index. Third place, tied with GPT-5.6 Sol Max, behind Anthropic's two leaders.
Why it matters
AI is becoming something you rent by the meter, and this launch shows how the meter really works. SpaceXAI says the sticker rate is $2 per million input tokens in August, yet past 200,000 tokens the price doubles versus the sticker, which means a long job pays the higher rate on every token.
That line sits exactly where the model's pitch lives, because long agent jobs are the work it is built to run. Artificial Analysis measured $0.84 per completed task in August, worse than the older Grok 4.5, which means the newest model is not automatically the cheapest way to finish work. Taken together, the numbers suggest the real benchmark is shifting from leaderboard scores to cost per finished job.
What to watch
If production teams report lower total bills per finished job, the pitch could hold; if long jobs keep crossing the doubling line, buyers would learn to split work or shop elsewhere.
Whether OpenAI or Anthropic answer on price is not yet known. Regulator files on the vendor remain open in the UK and Europe, and compliance teams read those files before they buy.
Reader comments
Not signed in yet — hit Post and we'll finish it together