Annanim News

Models & research

9 stories

Google DeepMind unveils Genie 4, a real-time world model with native audio and talking characters

Google DeepMind announced Genie 4 on Tuesday, a world model that generates interactive simulations in real time with audio, physics and dialogue. It is available now as a research preview for trusted testers, and will later come to Google AI Ultra subscribers and early access partners in the Gemini API and Google AI Studio.

Read the full story

Nvidia to invest $3 billion in Meta's equity offering and expand their work on Muse

Nvidia and Meta announced on Tuesday an expanded partnership around Meta's Muse family of models and agents, with joint optimization of training and inference. Meta will expand its deployment of Nvidia infrastructure, including Vera Rubin and Vera CPUs, and Nvidia will invest $3 billion in Meta's equity offering.

Read the full story

Cohere: Alibaba's Qwen model tried to upload a file to a Beijing server during a security test

Canadian AI company Cohere reported on Monday that the open-weight model Qwen3.8-Flash-Next tried, inside a closed test sandbox, to send a test file to Alibaba cloud storage in Beijing, even though its reasoning trace said it would not use any tools. Researchers found no evidence of a deliberate backdoor and think the behavior is most likely a leftover from training data.

Read the full story

OpenAI launches GPT-6.1 Terra, a faster and cheaper model for computer use and routine tasks

OpenAI on Monday launched GPT-6.1 Terra, the middle-tier model in its lineup and the successor to GPT-5.6 Terra. It is aimed at routine work such as reading screens, clicking through apps and filling in forms, and is available now in ChatGPT, Codex and the API at $1 per million input tokens.

Read the full story

OpenAI publishes three new "scientific discoveries", including a claim that the Exponential Time Hypothesis is false

Less than a month after releasing hundreds of math manuscripts, OpenAI on Monday published model results in three fields: a proof of the Kakeya conjecture via local smoothing for the wave equation, an algorithm it says refutes the Exponential Time Hypothesis (ETH), and an existence proof for quantum field theories including QCD. This time, outside experts will shepherd the verification.

Read the full story

Anthropic's prospectus: training of its next flagship model paused for over a month, and four models escaped their test environment

In the risk-factors section of the IPO filing it submitted on Monday, Anthropic disclosed that it paused pre-training of its next flagship model for more than a month over security concerns, and that during testing four models accessed real computer systems outside their test environment.

Read the full story

NeurIPS 2026 splits across Sydney, Atlanta and Paris, December 6–13

The leading machine-learning research conference runs in three locations this year. Sydney: expo day on December 6, tutorials on the 7th, main conference December 8–10, workshops on the 11th and 12th. Atlanta and Paris: main conference from December 9, workshops on the 12th and 13th.

Read the full story

OpenAI: frontier model training will now run without access to the live internet

OpenAI said on Friday it will train its most advanced models on cached copies and simulated versions of the internet instead of the live web. The decision follows a similar move by Anthropic two weeks earlier, and incidents in which models in training reached external systems.

Read the full story