Mistral previews its largest model
Mistral has opened a public preview of Mistral Large 4, a multimodal model built for coding, tool-using agents, image understanding, and professional documents. The company says it can create and edit complex spreadsheets and documents, while its demonstrations include retrieving evidence from PDFs and inspecting technical drawings.
The preview API is available now in Mistral Studio. Model weights are due by the end of October, opening the way for organizations to run it on their own infrastructure.
Search across media without sending it away
Google’s EmbeddingGemma 2 maps text, code, images, video, and audio into a shared format that software can search. It runs on local hardware, enabling private or offline tools that find a video clip from a voice memo or search audio recordings with a text query.
The 740-million-parameter model has an Apache 2.0 license, and its weights are available through Hugging Face and Kaggle. Google also provides demos for searching a media library and locating moments inside videos.
Also
- OpenAI publishes AI-generated math results The repository includes computer-checkable Lean proofs, reasoning summaries, attempted-problem statistics, and estimates of the compute used.
- Turn a described music tool into a studio plugin Google Flow Music lets producers describe custom instruments or effects and export them as VST3 or AU plugins for digital audio workstations.
- Anthropic expands vetted access for cyber work Three tiers give qualifying security professionals different levels of access to advanced capabilities and reduced blocking, with verification and controls increasing for higher-risk work.
- Claude builds a playable music artifact Simon Willison shares the prompt and result from asking Claude Opus 5.5 to design a text music format, compose game music, and build an artifact that plays it.
- When an AI gadget listens but does not save audio Victoria Song examines whether transcripts and summaries should count as recordings even when a device deletes the raw sound or video.
Pass it on