DeepSeek - Page 26
Stories from DeepSeek come first, newest first, then stories from connected topics (China, Open Source, AI Agents, AI, AI Models, Cybersecurity), each labelled with where it comes from. This view is not indexed by search engines.
Latest
Alibaba’s MAI-UI Team Open-Sources OpenGCUA for Generalist Computer-Use Agents
The MAI-UI team at Alibaba Token Hub open-sources OpenGCUA. Authors claim +19.6 mean partial score on OSWorld-v2 offline; harness and SDK are preliminary.
Anthropic Says Claude Traced NIST ECDSA Curve Seeds to Timestamped Random Output
Anthropic says Claude found NIST's ECDSA curve seeds came from lrand48 seeded with Unix timestamps. It is not a cryptographic break; no action is needed.
Epoch AI Says Open-Weight Models Trail Closed Ones on Cyber Skills by 6-7 Months
Epoch AI estimates open-weight models trail closed ones on cyber skills by 6-7 months, with Z.ai's GLM-5.3 on top, but its score rests on three results.
Goodfire Essay Claims Classic Superposition Evidence Was Misread
Goodfire’s Nick Cammarata argues neural nets are already mostly clean and classic superposition evidence was misread. Informal claim; critics push back.
Microsoft and NVIDIA Publish a Risk Guide for Open-Weight Models in Enterprise
Microsoft and NVIDIA's joint paper on open-weight model risk sets out five risk categories and argues openness alone does not decide risk. More to come.
Elorian’s Glimpse Beats GPT-6 Astra on 4 of 11 of Its Own Tests
Elorian's first model, Glimpse, reasons with images at $0.60 per million input tokens. Its own tests put it ahead of GPT-6 Astra on four of 11 benchmarks.
Vercel Says Jev Was the Fastest-Adopted Model on AI Gateway
Vercel says TypeSafe’s Jev reached 35% of paid AI Gateway teams in two weeks—the fastest adoption in gateway history. Gateway metrics only.
CERN and Mistral Sign Intent to Build a Large Physics Model
CERN and Mistral signed a statement of intent to develop a Large Physics Model for collider physics. The model is not built yet; no timeline disclosed.
Krea Launches Krea 3 Image Editing Model for Sub-8-Second 2K Edits
Krea launched Krea 3 for image editing: under 8s at native 2K and $0.06 per image, the company says. Speed and price figures are Krea’s own.
Reflection Releases Beam’s Open Weights on Hugging Face Under Apache 2.0
Reflection has published Beam's weights on Hugging Face: a 501B mixture-of-experts model, 23B active, with a 1M-token context and an Apache 2.0 licence.
Latest on TSN
Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.