Plus: Gemini 3.7 Flash ships, OpenAI quietly disbands its safety team, and DeepSeek's harness breaks GitHub records
Sponsored by
Get found in AI search before your competitors do. Genrank tracks the prompts, mentions, citations, and competitors shaping visibility across ChatGPT and other AI search engines. Start your free trial and see where your brand ranks.
Track your AI visibility with Genrank. Start your free trialThis week was the kind of week that makes you double-check the date on the headline. A rocket company now owns your code editor, a payments company is buying the router that sits in front of every model, Meta decided openness is its personality again, and OpenAI quietly folded up the team whose entire job was asking "what if this goes wrong?"
Today, we've got SpaceX closing its Cursor acquisition, Stripe's reported $7B+ swoop on OpenRouter, Meta's Apache-licensed Muse Glimmer, Google's Gemini 3.7 Flash, and the open-source agent harness that just became the fastest-rising repo in GitHub history.
On this day in 2025…
The AI world was still catching its breath from OpenAI's GPT-5 launch earlier that month — a model trained entirely on Nvidia's last-generation Hopper chips, before Blackwell had even shipped at scale. Twelve months later, the frontier has moved from "which model is smartest" to "who owns the scaffolding around it." Keep that in mind as you read the rest of this issue.
🚀 SpaceX Officially Owns Cursor
The deal is done: AI coding startup Cursor is now formally part of SpaceX. It's one of the strangest pairings in recent tech history — a launch provider absorbing the editor that defined the vibe-coding era — and it raises an obvious question for every developer with Cursor in their toolchain: whose roadmap are you on now?
💳 Stripe Is Buying OpenRouter for $7B+
OpenRouter's CEO has described the company as "Stripe for AI." Stripe apparently agreed and decided to just buy it. The reported $7 billion-plus deal puts the world's biggest payments company in control of the gateway that thousands of apps use to route requests across model providers — inference metering as a payments primitive.
🪟 Meta Returns to Open Source with Muse Glimmer
Meta released Muse Glimmer, a 30B-parameter, Apache 2.0-licensed model with a 128K context window, distilled from its closed Muse Spark and tuned for agentic work on a single consumer GPU. It leads Gemma 4 and Qwen3.6-27B on several agent benchmarks (MCP Atlas 75.5, SWE-Bench Pro 51.2) — and it arrived with a 14-page Zuckerberg essay arguing superintelligence should be distributed, not centralized.
⚡ Google Ships Gemini 3.7 Flash — But Still No Pro
Google launched Gemini 3.7 Flash for coding and automated business tasks, rolling it out immediately to Gemini Spark subscribers in 160+ countries. What it didn't announce: a date for the flagship 3.5 Pro, reportedly delayed two months after internal testing showed it trailing rivals on coding.
🛑 OpenAI Disbanded Its Preparedness Team
Per the Financial Times, OpenAI dissolved the team responsible for assessing whether models pose serious risks, redistributing responsibility into existing bio and cyber teams. The timing is uncomfortable: this is the same company whose autonomous agent escaped a sandbox in July and hacked Hugging Face.
Stop Guessing What People Ask AI
Your brand can rank well on Google and still be invisible when customers ask ChatGPT, Perplexity, Gemini, and other AI search engines for recommendations.
GenRank helps you see the prompts that drive AI visibility, so you can stop guessing and start optimizing for real search behavior.
GenRank helps you discover the prompts your audience is actually asking and identify where your brand gets mentioned, cited, or overlooked.
The goal isn't to track thousands of prompts. It's to find the right prompts that influence recommendations, buying decisions, and brand visibility.
Research the prompts. Track your visibility. Improve your position in AI search.
🧩 Cordis is the quiet TypeScript meta-framework doing the heavy lifting underneath DeepSeek Harness. Billed as "a meta-framework of spatiotemporal composability," it's a plugin kernel where every capability loads as the same kind of object — no privileged core to work around. MIT licensed, 4.7k stars, and worth a read if you're designing anything modular and long-running.
🥁 Eigendrum is the palate cleanser your week needs. Draw any shape, and it solves the Laplacian eigenvalue problem with finite element analysis to synthesize the sound that shape would actually make as a drum — nodal lines visualized and all. Vanilla JS, Web Audio API, custom sparse solvers, zero runtime dependencies, and it works offline from a local index.html.
Coming in Hot
The AI Search Platform for marketing teams.
Free AI humanizer that bypasses AI detectors
Intelligent canvas for AI creative studios
Free AI agents for workflow automation
Amazon seller & brand intelligence platform
AI-powered data enrichment for sales teams
AI-powered social media scheduling and marketing
AI video creation for musicians and creators
AI music video generator from any song
B2B data platform and APIs for sales and AI agents
Submit your AI tool to the #1 AI tools marketplace.
What it does: HumanizeAIText.io is a free AI text humanizer that rewrites content from ChatGPT, Claude, Gemini, and other AI writing tools to make it sound more natural while preserving the original meaning, facts, and structure. It supports 40+ languages, multiple writing modes, AI detection, and up to 1,000 words per request with no signup required.
Why It's Worth Your Time
Pricing
What it does: Astorie is an AI creative canvas that brings 100+ AI models for image, video, audio, music, and 3D creation into one visual workspace. It lets creators generate, edit, compare, and connect outputs through node-based workflows, making it easier to build everything from product visuals and social content to multi-shot AI videos and complete creative projects.
Why It's Worth Your Time
Pricing
🌊 A Two-Month-Old Startup Raised $1.1 Billion
River AI, founded by xAI co-founder Igor Babuschkin, pulled a $1.1B seed/Series A led by General Catalyst and AMP PBC, with Nvidia, AMD Ventures, Y Combinator and Temasek along for the ride. The pitch: any enterprise can finish a complex RL run in 15–20 minutes with no infra team.
🖊️ How Claude's Watermarks Actually Work
Anthropic is applying watermarking at the model level for everything released after August 2, using C2PA for files. Because the mark lives in the text itself, it survives copy-paste and some editing — which raises very real questions for anyone shipping AI-written code.
🧠 Anthropic on What Breaks in Multi-Agent Systems
A genuinely useful research write-up on the emerging patterns — and recurring failure modes — of multi-agent architectures. Required reading before you spin up your fifth subagent.
That's it for this week. A rocket company owns a code editor, a payments company owns the model router, and the fastest-growing repo in GitHub history is a plugin system. Normal stuff.
Hope you found something useful in here. If you did, send it to someone who'd appreciate it. 🎉
Follow us on Threads, X, and Instagram for more AI updates.
Signing off,
— Somi AI
