Key Takeaways
- Replit launched a free, OpenAI-powered version of its platform (GPT-5.6 Luna) that is a real buildable tier, not a trial; the company claims enterprise business grew more than 10x before this consumer expansion.
- OpenAI’s first NVIDIA Vera Rubin racks are installed and running the training stack, marking a concrete next step in scaling pretraining for frontier AI.
- Perplexity’s Agent API exposes 41 frontier models from 9 providers through one endpoint, including web search, financial search, scraping, and sandboxed code execution tools.
- Orbital AI compute startup Starcloud raised $250M at a $2.3B valuation, with Alexis Ohanian participating in the round.
- Agnost AI says its first model trained from production agent traces boosted task success by 22.9%, cut latency by 90.2%, and reduced cost by 94.5% for an early customer.
- Shopify’s CEO, Stripe’s CEO, Vercel’s CEO, and Basecamp’s founder are collectively going all-in on making “the year of Linux desktop” real.
1. AI Models and Developer Platforms
- Replit and OpenAI announced a partnership that makes OpenAI’s latest model, GPT-5.6 Luna, the engine of Replit’s new free mode. Masad says it is a real free tier for building, not a trial, and that enterprise business grew more than 10x after his “vibe coding goes to zero” thesis; he also shared examples of users shipping paid products quickly after launch. — via 1 2 3
- Perplexity Agent API: According to Perplexity CEO Aravind Srinivas, a complete AI developer platform requires multi-model access plus production deployment tools; the Agent API provides 41 frontier models from 9 providers through a single endpoint, with built-in web search, financial search, crawling, and sandbox code execution. — via 1
- Stripe is opening early access to a Provisioning API for developer and “vibe code” platforms, letting agents create accounts at providers like Supabase, OpenRouter, and Cloudflare through code rather than a browser. — via 1
- Patrick Collison calls new model Ox Alpha “very impressive,” citing a 1M token context window, text/image/video input, and suitability for efficient coding and agentic work. — via 1
- Twitter co-founder jack says he was forced to switch from Claude Code to Codex and found Codex better, faster, and cheaper; the trigger was Fable using up all of his Claude tokens. — via 1
2. AI Infrastructure and Compute
- OpenAI CEO Sam Altman shared an infrastructure milestone: the first NVIDIA Vera Rubin racks have arrived and are running the training stack, providing compute for scaling next-generation frontier AI pretraining. — via 1
- Starcloud, an orbital AI compute company, raised $250M at a $2.3B valuation to expand on-orbit AI compute capacity; Alexis Ohanian says he invested in the round again. — via 1
- Vercel released version 0.0.5 of an unspecified tool now small enough to fit on two floppy disks (xz-compressed) and going live tomorrow; Vercel CLI also gains Container Registry management via vercel vcr build docker --push. — via 1 2
3. AI Startup and Research Milestones
- Agnost AI released its first model, agnost-***-0.1, built on the idea of using agent trace logs for training rather than just debugging. The first customer trained on production traces saw task success +22.9%, latency -90.2%, and cost -94.5%. — via 1 2
- Patrick Collison announced the 2026 Virtual Cell Challenge, a $100k prize sponsored by NVIDIA and others, modeled on CASP, to create standardized benchmarks for virtual cell machine-learning models and speed progress. — via 1
- Alexis Ohanian invested in an independent studio with Naruto rights-holder approval to develop the franchise’s first digital card game. — via 1
4. Industry Signals and Controversies
- A growing group of founders is backing “the year of Linux desktop”: Shopify CEO tobi lutke says all conditions are ready and he’s all-in; Stripe CEO Patrick Collison and indie hacker levelsio amplified the call; Basecamp’s Jason Fried says he and others each put $1M into the initiative and support Omacom Foundation fundraising for Linux developer vaxryy. — via 1 2 3 4 5
- Naval argues that front-runner AI labs have an incentive to amplify fear to justify closed releases, while Chinese labs have consistently advocated openness and benefit all humanity, leaving smaller players with little choice but to side with China. This is a disputed/unverified framing rather than an established fact. — via 1
- Data-center politics have become a contested narrative: Alexis Ohanian says AI and energy are economic productivity and growth engines and that being anti-data-center is anti-labor, while Garry Tan calls the data-center discussion “a documented psyop” and cites allegations of a coordinated foreign influence operation using CCP-affiliated media, a Shanghai Marxist nonprofit network, and over $2B in dark money to penetrate U.S. advocacy. These claims are disputed/unverified and reflect conflicting viewpoints. — via 1 2 3
- Garry Tan and Paul Graham both argue that capitalism is won by making customers’ lives better, not by squeezing them: squeezing yields roughly 2x, while creating new things can yield 10x. — via 1 2
