Replit Opens a Free Mode, Powered by GPT-5.6 Luna

A new zero-cost tier puts OpenAI's low-cost Luna in front of every builder — and Replit says the era of expensive coding is over.

Replit launched Free Mode, a new tier powered by OpenAI's GPT-5.6 Luna, built to make creating with AI radically cheaper. The company's argument is blunt: agents made software cheaper, but they made coding expensive. Pairing Luna's efficiency with Replit's build, deploy and design tools is meant to flip that trade-off. Replit says Core and Pro users can now build up to 30× more for the same cost, and that it is working more closely with OpenAI than ever.

Early signals are bullish. Replit users have already used the platform to stand up high-value businesses and replace legacy software, and the company frames Free Mode as the on-ramp for anyone who has not yet felt AI's benefits. The launch lands as OpenAI pushes Luna as its answer to the industry's cost problem.

By @Replit · Product

OpenAI Keeps Its Zero-Data-Retention Promise for Frontier Models

OpenAI reaffirmed that its frontier models will continue to offer Zero Data Retention, and previewed a new safety system that flags risks across long, autonomous tasks. As agents take on longer, more consequential work, the company argues safety tooling must look beyond single turns and trace risks across related interactions. Sam Altman echoed the move with a two-word post backing business privacy.

Vercel Agent Joins Slack with Full Production Context

Vercel's agent can now be dropped into any Slack thread, bringing the full production context of your apps and agents. Mention @vercel to diagnose incidents, review pull requests, roll back deploys or update configuration — without leaving the conversation. The integration is in public beta for Pro and Enterprise plans.

Hyra Nudges a Decades-Old Math Bound Toward Optimal

Tencent's Hyra raised the universal lower bound for constant-width bodies from 0.380799w³ to 0.411040w³, reaching 97.9% of the conjectured Meissner optimum. The 3D Blaschke–Lebesgue result is one of four math advances the model has shipped since its last update.

Cursor Cloud Agents Pick Up Work and Hold a Goal

Cursor's cloud agents now take over tasks from events, hold a goal until it is met, and stay on course through long sessions. The update pushes the agent further toward trustworthy, multi-step autonomy.

Asana Finishes a "Five-Year" Test Migration in Two Weeks

With Codex, Asana migrated its frontend tests from Enzyme to React Testing Library in two calendar weeks — a project the team estimated would take five more years.