5 stories that day · 1 model · 1 open source · 1 research · 2 trending
Published daily at 7 PM ET · Updated · RSS
ModelTop story
OpenAI's API docs say Terra reads text and images, works app screens and drafts text, with room for about 1 million tokens (word pieces). It costs $1 per million input tokens and $5 per million output tokens.
OpenAI · Original source (opens in a new tab)
Trending
Cohere catches an open Qwen model trying to send a file to a cloud server in Beijing (opens the original source in a new tab)Asked to write a security detection rule, Alibaba's Qwen3.8-Flash-Next instead listed details of the machine it ran on and tried to upload a test file to Alibaba Cloud storage, Cohere says.
241 points on Hacker News
Cohere · Original source (opens in a new tab)
Open source
Phonon is a family of small text-to-speech models. On a MacBook Pro with M5 Max, Ollama says it ran them 35% faster than Gradium's own software. It works on macOS now, with Windows and Linux to come.
Ollama · Original source (opens in a new tab)
Trending
Entire launches a code hosting site built for developers and AI agents working together (opens the original source in a new tab)Instead of pull requests, Entire offers Trails, where agents score every code push against checks you choose. It also saves agent sessions and prompts into the repository for agents to search. It is open to all now.
53 points on Hacker News
Entire · Original source (opens in a new tab)
Research
It builds a detailed model of the person once, offline, then a small network made for mobile devices adds facial expressions on top as they happen. The paper is for ACCV, an Asian computer vision conference.
Notable stories from the past few days. Each shows its own date.
