AI lab announcements, 20–26 October 2025

38 announcements from 8 labs, first-party only

Google DeepMind

Google DeepMindNew model· 25 Oct

T5Gemma: A new collection of encoder-decoder Gemma models

Google DeepMind announces T5Gemma, a new collection of encoder-decoder large language models.

Google DeepMindNew model· 25 Oct

MedGemma: Our most capable open models for health AI development

Google DeepMind announces new multimodal models in the MedGemma collection, their most capable open models for health AI development.

Google DeepMindNew model· 25 Oct

Introducing Gemma 3n: The developer guide

Gemma 3n is designed for the developer community that helped shape Gemma.

Google DeepMindAvailability change· 25 Oct

Gemini 2.5 Flash-Lite is now ready for scaled production use

Gemini 2.5 Flash-Lite, previously in preview, is now stable and generally available, featuring a 1 million-token context window and multimodality.

Google DeepMind· 25 Oct

Behind “ANCESTRA”: combining Veo with live-action filmmaking

Google DeepMind partnered with Darren Aronofsky, Eliza McNitt and a team of more than 200 people to make a film using Veo and live-action filmmaking.

Google DeepMindNew model· 24 Oct

AlphaEarth Foundations helps map our planet in unprecedented detail

Google DeepMind has introduced a new AI model that combines petabytes of Earth observation data into a unified representation, aimed at improving global mapping and monitoring.

Google DeepMindCapability change· 24 Oct

Exploring the context of online images with Backstory

Google DeepMind announces Backstory, an experimental AI tool that helps people explore the context and origin of images seen online.

Google DeepMind· 24 Oct

Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad

The source text provides background on the International Mathematical Olympiad but does not contain any AI product or model announcement.

Google DeepMind· 24 Oct

Genie 3: A new frontier for world models

Google DeepMind announces Genie 3, a world model capable of generating consistent dynamic worlds that can be navigated in real time at 24 FPS and 720p resolution for a few minutes.

Google DeepMindNew model· 24 Oct

How AI is helping advance the science of bioacoustics to save endangered species

Google DeepMind announced a new Perch model that helps conservationists analyze audio faster to protect endangered species.

Google DeepMind· 24 Oct

Using AI to perceive the universe in greater depth

The blog post discusses using AI to perceive the universe in greater depth.

Google DeepMindCapability change· 24 Oct

Gemini achieves gold-medal level at the International Collegiate Programming Contest World Finals

Gemini 2.5 Deep Think achieved gold-medal-level performance at the ICPC World Finals, demonstrating a major advance in abstract problem solving.

Google DeepMindResearch paper· 24 Oct

Discovering new solutions to century-old problems in fluid dynamics

Google DeepMind introduces a new method to help mathematicians leverage AI techniques to solve long-standing challenges in mathematics, physics, and engineering.

Google DeepMindSafety or policy· 23 Oct

Strengthening our Frontier Safety Framework

Google DeepMind announces strengthening of the Frontier Safety Framework to identify and mitigate severe risks from advanced AI models.

Google DeepMind· 23 Oct

Gemini Robotics 1.5 brings AI agents into the physical world

The post announces that Google DeepMind is ushering in an era of physical agents, where robots can perceive, plan, think, use tools, and act to solve complex multi-step tasks.

Google DeepMindProduct launch· 23 Oct

Introducing CodeMender: an AI agent for code security

Google DeepMind introduces CodeMender, an AI agent for code security that uses advanced AI to fix critical software vulnerabilities.

OpenAI

OpenAIPartnership· 23 Oct

OpenAI acquires Software Applications Incorporated, maker of Sky

OpenAI acquired Software Applications Incorporated, the company behind Sky, a natural language interface for macOS.

OpenAI· 23 Oct

Consensus accelerates research with GPT-5 and Responses API

This is a customer story about Consensus using GPT-5 and the Responses API to build a multi-agent research assistant.

OpenAICapability change· 23 Oct

Work smarter with your company knowledge in ChatGPT

OpenAI introduced a new feature called Company Knowledge that integrates context from users' apps into ChatGPT to provide business-specific answers with citations, security, privacy, and admin…

OpenAI· 23 Oct

AI in South Korea—OpenAI’s Economic Blueprint

OpenAI published a blueprint for South Korea's AI economy.

OpenAIAvailability change· 22 Oct

The next chapter for UK sovereign AI

OpenAI announced a partnership with the UK Ministry of Justice to bring ChatGPT to civil servants and introduced UK data residency for ChatGPT Enterprise, ChatGPT Edu, and the API Platform.

OpenAI· 22 Oct

AI in Japan—OpenAI’s Japan Economic Blueprint

OpenAI published a document titled 'Japan Economic Blueprint' that describes how Japan can use AI to drive innovation, competitiveness, and sustainable growth.

OpenAIAvailability change· 21 Oct

Continue your ChatGPT experience beyond WhatsApp

OpenAI announces that ChatGPT will be removed from WhatsApp on January 15, 2026, and instructs users to link their accounts to continue conversations on other devices.

OpenAIProduct launch· 21 Oct

Introducing ChatGPT Atlas, the browser with ChatGPT built in

OpenAI announces ChatGPT Atlas, a browser with integrated ChatGPT, offering instant answers, summaries, and web help, with privacy controls, available for MacOS.

Hugging Face announces LeRobot v0.4.0, a major upgrade to the open-source robot learning framework, featuring new dataset infrastructure (Datasets v3.0), dataset editing tools, support for LIBERO and…

Hugging FaceDeveloper tool· 23 Oct

Building the Open Agent Ecosystem Together: Introducing OpenEnv

Meta and Hugging Face are partnering to launch the OpenEnv Hub, a shared community hub for agentic environments. They are also releasing the OpenEnv 0.1 Spec (RFC) for community feedback.

Hugging FacePartnership· 22 Oct

Sentence Transformers is joining Hugging Face!

Sentence Transformers is transitioning from the UKP Lab at TU Darmstadt to Hugging Face, with Tom Aarsen continuing as maintainer. It remains an open-source project under Apache 2.0.

Hugging FacePartnership· 22 Oct

Hugging Face and VirusTotal collaborate to strengthen AI security

Hugging Face announces a partnership with VirusTotal to continuously scan all public model and dataset repositories on the Hugging Face Hub for malicious files, enhancing AI security.

Hugging FaceCapability change· 21 Oct

Unlock the power of images with AI Sheets

Hugging Face AI Sheets, an open-source spreadsheet tool for working with AI models, adds vision capabilities.

Hugging Face· 21 Oct

Supercharge your OCR Pipelines with Open Models

Hugging Face published a guide to open-weight OCR models, covering model capabilities, output formats, and tools for running models locally or remotely.

Google Research

Google ResearchModel update· 23 Oct

Google Earth AI: Unlocking geospatial insights with foundation models and cross-modal reasoning

Google Research announces new versions of the Earth AI family of models, specifically new Imagery and Population foundation models, with state-of-the-art performance.

Google ResearchResearch paper· 22 Oct

A verifiable quantum advantage

Google Quantum AI introduces a new quantum algorithm called Quantum Echoes for measuring Out-of-Time-Order Correlators (OTOCs), demonstrating a verifiable quantum advantage on the Willow quantum…

Google ResearchResearch paper· 20 Oct

A picture's worth a thousand (private) words: Hierarchical generation of coherent synthetic photo albums

Google Research introduces a method for generating differentially private synthetic photo albums that uses an intermediate text representation and produces the albums hierarchically, preserving…

Google ResearchStory of the weekResearch paper· 20 Oct

Teaching Gemini to spot exploding stars with just a few examples

Google Research, in collaboration with the University of Oxford, published a paper in Nature Astronomy demonstrating that Google's Gemini model, using few-shot learning with only 15 annotated…

LG AI Research

LG AI Research· 21 Oct

LG AI Tech CONNECT - AI Product: Connecting the Present and Future of AI

The source text reports on an internal and external networking event hosted by LG AI Research, featuring presentations on EXAONE-powered products and the company's vision, with no new product or…

Mistral AI

Mistral AIProduct launch· 24 Oct

Introducing Mistral AI Studio.

Mistral AI announced the launch of Mistral AI Studio, a new product.

Replicate

ReplicateAvailability change· 21 Oct

Extract text from documents and images with Datalab Marker and OCR

Datalab's Marker and OCR models for document parsing and text extraction are now available on Replicate.

Together AI

Together AIResearch paper· 22 Oct

Large Reasoning Models Fail to Follow Instructions During Reasoning: A Benchmark Study

This blog post introduces ReasonIF, a benchmark for evaluating instruction-following in reasoning traces of large reasoning models (LRMs).