Grok gets smarter, Muse beats ChatGPT's launch, Amazon slams the door

On September 21, SpaceXAI (the company formerly known as xAI) released Grok 4.7, built on a larger base model than Grok 4.6 with a longer reinforcement-learning run weighted toward tasks that take many hours to finish. Price held steady at $2 per million input tokens and $6 per million output tokens, the same as Grok 4.6, with a faster variant at double the speed for double the price.

On the company's own CursorBench 4.0 coding benchmark, Grok 4.7 scored 46.3%, ahead of Grok 4.6's 40.4% and GPT-5.6 Sol Max's 41.7%, though still behind Fable 5.1 Max's 51.8%. SpaceXAI also says it is the most jailbreak-resistant model it has tested, topping LatchBio's biosafety benchmark at 62.4% and letting only 3.3% of risky cyber prompts through on HackerBench v0.3. It is live now in Cursor, Grok Build, the Grok API and assorted third-party coding tools.

Meta's Muse personal AI agent launched September 8. By September 18 it had reached No. 1 on Apple's US App Store with more than 730,000 downloads in ten days, according to Sensor Tower data reported by Yahoo Finance, overtaking ChatGPT, Gemini and Claude on the free chart. Wall Street noticed: JPMorgan upgraded Meta stock and raised its price target from $640 to $820.

TechCrunch, citing analytics firm Apptopia, ran the apples-to-apples version: in the first 12 days on iOS in the US and Canada, Muse pulled 1.8 million downloads against ChatGPT's 1.3 million at the same stage after its own 2023 launch, with daily active users running about 642,000 versus 231,000.

The celebration did not last. On September 20, Amazon cut Muse off from shopping on Amazon.com, saying the agent never disclosed itself while browsing and appeared to capture customer credentials without permission. Amazon has now blocked ChatGPT, Claude, Copilot, Perplexity and Grok from shopping on its site too, while its own Alexa keeps a standing invitation to buy elsewhere.

Marketing insight: Every big retailer is now deciding, on its own, which AI agents get to shop there. If agentic commerce is anywhere on your roadmap, you need a platform-by-platform plan, not one universal AI checkout strategy.

On September 21, OpenAI argued that the United States should lead an international push to set technical standards for frontier AI, with particular focus on what it calls recursive self-improvement: AI systems upgrading their own capabilities with less human oversight over time. The company says it is not pursuing that today and will not "unless and until it can be done safely."

The timing is not subtle. Altman is scheduled to brief the UN Security Council on AI safety this Wednesday, September 23, the council's first AI-focused session since 2023, according to Benzinga.

Line these up and a pattern shows up. A model got cheaper and stronger without asking permission. An app agent got popular enough to threaten a trillion-dollar company's storefront rules. A lab asked a government to referee before things move faster than anyone can watch. None of these are chatbot stories anymore. They are stories about who controls access, data and trust once software starts acting on its own.

For your business, the lesson holds steady: the tools are getting cheap and capable fast, but the fights over where AI is allowed to operate, and on whose terms, are just getting started. Read the terms of service before you build your roadmap around any one of them.

If you want help sorting which of this week's shifts actually change how your customers find, trust and buy from you, that is the exact conversation our founders like having. Talk to them below. See you in the next edition.