The Archive

T5Gemma: Google releases new open-weights encoder-decoder LLM variants in Gemma family.

MedGemma: Google announces multimodal open-weights models for healthcare AI development.

Gemma 3n developer release with optimized tooling for open-weights model community.

Gemini 2.5 Flash-Lite reaches general availability with 1M context, multimodality, production-grade performance.

Google showcases video generation via Veo in Darren Aronofsky film collaboration.

AlphaEarth Foundations model processes petabytes of satellite data for unified planetary mapping.

Mistral AI Studio is a production platform for deploying and managing AI applications with enterprise integration.

Backstory: experimental AI tool for verifying image context and origin online.

Gemini with Deep Think achieves gold-medal performance at International Mathematical Olympiad.

Google DeepMind releases Aeneas, a model for contextualizing and restoring fragmentary ancient inscriptions for historical research.

Genie 3 generates navigable 720p dynamic worlds at 24fps with multi-minute consistency, advancing world model capabilities.

Google DeepMind's Perch model accelerates bioacoustics analysis for endangered species conservation, including Hawaiian honeycreepers and coral reef monitoring.

Headline-only announcement about AI applications in astronomy; no substantive detail provided.

Gemini 2.5 Deep Think achieves gold-medal performance at ICPC World Finals, demonstrating major advance in abstract reasoning and competitive programming.

Google DeepMind reports a method enabling AI-assisted solutions to long-standing problems in fluid dynamics, mathematics, and engineering.

Google DeepMind strengthens Frontier Safety Framework to identify and mitigate severe risks from advanced AI models.

Gemini Robotics 1.5 enables physical AI agents to perceive, plan, and execute multi-step tasks with tool use in real environments.

CodeMender, an AI agent that automatically identifies and fixes critical software vulnerabilities in code.

Anthropic establishes Seoul office as third Asia-Pacific location to expand regional presence.

Google DeepMind partners with Commonwealth Fusion Systems to apply AI to accelerate commercial fusion energy development.

Google DeepMind rolls out Deep Think reasoning capability in Gemini app for Ultra subscribers; mathematicians get IMO competition model access.

Google DeepMind open-sources Game Arena, a benchmark platform for head-to-head evaluation of frontier AI models in competitive environments.

Google DeepMind releases Gemma 3 270M, a compact 270M-parameter model optimized for efficient inference and resource-constrained deployment.

Gemini app gains enhanced native image editing capabilities with improved transformation and control features.

Anthropic expands use of Google Cloud TPUs and services for infrastructure scaling.

Google DeepMind introduces VaultGemma, claiming the highest-capacity language model trained from scratch with differential privacy guarantees.

Google DeepMind unveils Gemini 2.5 Computer Use model in API preview; specialized for UI automation and agent-based workflows.

Google DeepMind updates Veo 3.1 video model with enhanced creative controls for video generation and editing.

Google DeepMind releases 27B foundation model for single-cell genomics analysis, derived from Gemma, enabling cancer therapy pathway discovery.