Researchers find 81 of 134 AI models fail terrorism-safety benchmark
Every safeguard-stripped model tested failed. But the benchmark measures willingness to answer—not whether someone could act on the advice.
Superpower Daily4 min read
Models
25 stories
Fish Audio releases Drama 3 with script-directed voices and downloadable weights
Creators can direct delivery mid-sentence through the website, app and API. Editing individual words remains in testing, and the downloadable model carries commercial-use limits.
Superpower Daily4 min read
NEURA introduces robot AI built to combine sight, touch, sound and memory
Neuron AI is intended to work across five NEURA robot designs. A planned laundry demonstration will show its approach to multistep work, but independent performance comparisons remain pending.
Superpower Daily3 min read
Claude Opus 5.5 leads Vals AI’s visual bug-fixing test by nearly nine points
The evaluation combines screenshots with real JavaScript issues. A network-access diagnostic shows why letting agents search for published patches can sharply change the results.
Superpower Daily3 min read
Princeton’s Kothari says AI did most of a math advance, with humans steering
In comments published October 13, Kothari describes unusually limited human involvement in an earlier mathematical result—and a harder question about training future researchers.
Superpower Daily3 min read
Artificial Analysis adds a 3D-building benchmark, with four AI models tied at the top
AA-3D measures interactive scenes through blind human votes. Its launch results link early visual inspection with stronger scores, while scene costs vary sharply.
Superpower Daily4 min read
Anthropic says stronger Claude defenses cut extracted reasoning by 90–99%
The safeguards restrict how reasoning moves between accounts and models. Anthropic reports sharp reductions in extraction, but acknowledges mistaken blocks, integration changes and charges for some rejected requests.
Superpower Daily4 min read
Open Weight Models That Actually Hold Up on Real Coding Work
We ran open weight models on real pull requests. GLM-5.3 Flash landed 19 of 20 fixes, same as Opus in Claude Code, for $1.64 instead of $41.34.
Zentask Blog12 min read
The model that didn't exist, so you made it yourself
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Hugging Face Blog9 min read
China's DeepSeek, peers launch 16 AI models in month despite Anthropic warning
GUANGZHOU/TOKYO -- China's artificial intelligence development drive is showing no signs of slowing, with DeepSeek, Xiaomi and others rolling out mode
Nikkei Asia Tech1 min read
The Desert Ant CLI: on-device AI models for you and your agents
Run models on your own machine from the terminal, or hand them to your coding agent. None of your files leave your machine.
Desert Ant Labs Blog4 min read
GPT-6 Astra
OpenAI shipped GPT-6 Astra on September 3, 2026, its first flagship release since GPT-5.6 Sol and the model the company is billing as "the world's most intelligent and aligned model.
Awesome Agents Newsletter Blog10 min read
Bonsai 1.7B
Bonsai 1.7B has been the quiet entry in PrismML's lineup since March, overshadowed by the flagship Bonsai 27B that made headlines for fitting a 27B-class model on an iPhone.
Awesome Agents Newsletter Blog8 min read
Auto mode: an experiment in letting a 250 ms model pick the model for every message
We put Jev in front of every message in OpenChamber so a cheap classifier picks the model. It worked great until I typed "do it". Here's what we experimented with, where it pays off, where it doesn't yet, and what we learned about asking a model the right question.
OpenChamber Blog7 min readJev by TypeSafe AI: 193× in the headlines, 7× in user reports
We analysed 12,759 tweets about Jev's launch to separate TypeSafe AI's speed and cost claims from users' reported results and see where a decision model fits.
OpenChamber Blog17 min read
From 10-minute recording to edited content in 14 seconds
How we rebuilt Detail's Auto Edit with six Desert Ant models running entirely on-device: lower cloud costs, simpler code, and faster, more predictable results.
Desert Ant Labs Blog6 min read
Rebuilding AUTOMATIC1111 with Gradio Workflow
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Hugging Face Blog9 min read
On-device intelligence for every product
We're building a frontier lab for on-device AI. Small, specialized models for audio, vision, and text, faster than the cloud and free.
Desert Ant Labs Blog6 min read
Showing 18 of 25 stories
Load more storiesAiByts Weekly Brief
Get the top AI news and tools straight to your inbox every week.
Weekly · AI news
Join 15,000+ AI enthusiasts