Underdog released a browser game built on Woof, a 31.7 MB text-to-speech model with 31 million parameters that the company says can run offline on most devices. The demo illustrates in-browser speech, music generation and an open download.
OrgSync is a role-aware company knowledge system that feeds Claude, ChatGPT and Codex with approved, cited decisions and docs. Founder Kaustubh Dhotre says it centralizes goals and weekly reviews, restricts access by role, and launched in 2026 with a free week trial.
The Open ASR Leaderboard added private Egyptian and Nigerian English splits through a collaboration with Liva AI, supplying conversational datasets of 1.34 and 1.55 hours. The change expands accent coverage and updates the leaderboard’s accent breakdown based on Kachru’s Three Circles.
Translucent AI offers an agentic system that chooses the next physical test for hardware teams, records decisions and limits solver runs. In simulated battery trials it reported a 33.8-minute 10–80% charge versus 42.6 for Bayesian optimisation; no physical cells were tested.
BuildFerry launches a service that registers tester iPhones and re-signs .ipa builds in the cloud, removing UDID collection, Macs and TestFlight for ad hoc distribution. Free for one app; Solo $12/mo and Pro $29/mo. iOS only, launched in 2026.
Kivo is a free Mac app that inserts cleaned dictation text into any app and keeps audio on the device. The maker, Gokulakrishnan S, says it is an early English-only preview requiring a one-time 460 MB model download and an unsigned app approval.
refern has been rebuilt as a desktop app that keeps images, videos and saved pages alongside working files. The maker says its 2024 web app reached 23,000 users; the new desktop version runs on Windows, macOS and Linux and is $35 with a launch-week discount.
Cult.fit standardized its engineering workflow on Codex and expanded the platform to non‑engineering teams, enabling internal app building, data self‑service and voice‑first features like an AI Trainer. The company reports cost savings and lower support costs.
Merge deploys GPT‑6 Astra, Codex and ChatGPT across product development, integrations and internal workflows, CEO Shensi Ding says. Merge reports tool-call reliability of 99.9% versus ~60% on public MCP servers and says an AI “dinner bot” cut planning to about 30 minutes per event.
NVIDIA released four open-weight Cosmos3-Nano checkpoints—Cosmos3-Nano-Transfer-Auto, Cosmos3-Nano-Sim-Auto, Cosmos3-Nano-Sim-Transfer and Cosmos3-Nano-Sim-Bimanual—for multi-sensor driving and bimanual robot video prediction. Each model is a 16B-parameter, post-trained checkpoint distributed under