The Anthropic SDK for Python has released version 1.14.0, which includes new features such as adding execution_identity to Managed Agents and native support for OpenTelemetry tracing. The update also fixes several bugs, including renaming the telemetry destination test response model and closing dropped streams without waiting for the collector. Additionally, the documentation has been updated to clarify when custom tool results can be sent and to document the scope required to assign the managed role.

Google DeepMind and GovTech Singapore have collaborated to implement on-device AI for scam detection. By utilizing lightweight Gemma models, the system can intercept scam attempts in private chats without sending any data to the cloud. This approach aims to protect users from conversational fraud while maintaining strict data privacy.

Google DeepMind has launched Gemini 4 Argon, a frontier model designed for high-precision tasks in domains such as coding, finance, legal, and science. The model is available via API in AI Studio and the Gemini platform, and for subscribers through Google Antigravity and the Gemini app. Early implementations show significant performance gains in codebase refactoring, quantum chemistry optimization, and accelerating open-source model inference.

The Anthropic SDK Python version 1.13.0 has been released, featuring additions to the API such as types for unified analytics metrics and enhancements to Managed Agents. This update also includes internal improvements like naming the default API version header value. These changes are part of 32 commits made to the main branch since the last release.

Google DeepMind has launched EmbeddingGemma 2, a lightweight multimodal embedding model released under the Apache 2.0 license. Built on the Gemma 4 architecture, the 740M parameter model maps text, code, images, audio, and video into a unified embedding space. It is specifically optimized for on-device inference, supporting privacy-first RAG pipelines and local semantic search.

The latest release of the mistral-inference library, version 1.6.0, introduces support for the Mistral Small 3.1 model. This update includes minor code cleanups and documentation adjustments.

Mistral AI has released version 1.4.0 of the mistral-inference repository. This update introduces support for the Pixtral-12B-2409 model, enabling users to run multimodal inference via Python or the command line. The update allows the model to process text prompts alongside one or more images.