Latest in AI
Every brief, newest first. Each one links to its primary source.
-
Agents can copy how an app looks. Finishing it is another matter
A new benchmark gives the leading model 58.1% overall but only 2.8% of tasks passing every test, which says something about how agent scores…
-
Qwen’s new image model is open-weight, not open-licensed
Alibaba’s Qwen team has published a 7B image generation and editing model with native transparency, but the weights carry a research licence rather than…
-
UK and US link defence AI teams as an undersea drone fires a torpedo for the first time
The UK’s rapid AI taskforce will work with the US Chief Digital and AI Office, announced days after a UK uncrewed submarine fired a…
-
South-East Asia builds sovereign AI on NVIDIA’s open Nemotron models
Singapore, Malaysia, Vietnam and Thailand are fine-tuning NVIDIA’s open weights for local languages rather than training from scratch.
-
Anthropic prices Opus 5.5 down, and prints the benchmarks it loses
Opus 5.5 costs 40% less to run than Opus 5, and Anthropic’s own table shows GPT-6 Astra ahead on two of its eight benchmarks.
-
Google, Nvidia and Emerald AI launch an alliance on grid-flexible data centres
A new coalition wants data centres to adjust their power draw in response to. The structure and intent are set out, but no figures,…
-
Google ships Gemini 3.8 Live for real-time voice agents
Google’s new Gemini 3.8 Live audio models complete 35.1% of Sierra’s τ-Voice-banking tasks, the figure that says most about where voice agents actually are.
-
OpenAI publishes a youth safety blueprint for Australia
A six-pillar policy document aimed at Australian regulators that restates protections OpenAI has already shipped and adds no dated commitments.
-
Google puts generated classroom simulations in front of teachers
Google Research has released a library of more than 30 AI-generated STEM simulations for teachers, backed so far by a study of twelve of…
-
AWS rewrites its agent runtime and changes how memory is billed
A new platform version of Amazon Bedrock AgentCore reclaims memory during a session rather than billing peak allocation for its whole life.
