Mistral, Moderation, Medicare

By the Orchestra editors · 4 min read

TL;DR

Mistral drops a trillion-parameter open-weight model aimed at security work closed rivals refuse
Google ships EmbeddingGemma 2: a tiny multimodal embedding model that runs on-device
Deep tech coalition pushes Chalmers to rethink RDTI age limits
Anthropic tells Australian inquiry it's open to mandatory AI breach disclosure laws

AI · Models & Tools

Mistral Large 4 launches as Europe's biggest open-weight model

Mistral released Large 4, a trillion-parameter multimodal model trained entirely on European infrastructure. It makes a big jump on independent benchmarks but still trails Claude, GPT-6 and top Chinese models. The pitch: cybersecurity work that closed US models refuse to do.

Google's EmbeddingGemma 2 runs multimodal RAG on-device

Google DeepMind released EmbeddingGemma 2, a 740M-parameter open model that embeds text, images, video, audio and code into vectors. It needs roughly 191 MB of RAM and outperforms some models twice its size, making fully offline RAG pipelines practical when paired with a small Gemma model.

Musubi ships PolicyLM-1.7B for real-time content moderation

Musubi announced PolicyLM-1.7B, a lightweight open-weight decision model purpose-built for real-time content moderation. It sits in the growing category of small, task-specific models that teams can self-host rather than routing sensitive decisions through a general-purpose API.

Australian Tech

Deep tech coalition presses Chalmers to scrap RDTI age limit

A coalition of research groups, accelerators and incubators has written to Treasurer Jim Chalmers for a second time since the May budget, urging the government to drop or rework the proposed age cap on the R&D tax incentive that they say would undermine Australia's deep tech sector.

Anthropic open to Australian laws mandating AI breach disclosure

Appearing before the parliamentary AI inquiry, Anthropic said it has found no evidence its products accessed Australian government systems but would support laws requiring AI companies to disclose data breaches, drawing a pointed contrast with OpenAI's handling of the Medicare incident.

Advanced Navigation opens US manufacturing to meet defence rules

Sydney-based deep tech firm Advanced Navigation announced a new manufacturing operation in Alabama for its positioning, navigation and timing products, designed to satisfy US defence supply-chain requirements.