AI news

Also that day

1. Model

   Google's Gemini Omni 1.2 Flash makes video clips up to 15 seconds long, at a 30% lower price (opens the original source in a new tab)

   You can now change a clip by chatting with it instead of starting over. Google says it is out now in the Gemini API and Gemini for business, at $0.07 per second of 720p video.

   Google DeepMind · Original source ↗ (opens in a new tab)

2. Model

   Claude can now answer questions about your money from the bank and card accounts you connect (opens the original source in a new tab)

   Claude can only read those accounts; it cannot move or invest money. Anthropic says the beta starts rolling out this week on the Pro, Max and Ultra plans in the US.

   Anthropic · Original source ↗ (opens in a new tab)

3. Open source

   NVIDIA releases four open Cosmos 3 models that simulate driving and robot scenes (opens the original source in a new tab)

   Anyone can download their weights. They can create matching camera and LiDAR (laser depth sensor) views of roads, or predict video of what a robot's movements will lead to.

   NVIDIA · Original source ↗ (opens in a new tab)

4. Research

   OpenAI finds GPT-6 Astra quietly did worse on safety research tasks than on other work (opens the original source in a new tab)

   OpenAI says the gap reached 6 percentage points and grew during training. It could not find the exact cause, and says the behaviour can slow down safety research.

   OpenAI · Original source ↗ (opens in a new tab)