Key AI and LLM Happenings of the Past Week

1. Mistral Small 3.1

Mistral AI has released Mistral Small 3.1, an open-weight 24 billion parameter model that outperforms competitors such as Google's Gemma 3 and OpenAI's GPT-4o Mini in key benchmarks. The model supports multimodal inputs, handles a context of up to 128,000 tokens, and processes 150 tokens per second, making it highly efficient. This could accelerate the development of AI applications, particularly in the open-source sector.

2. Claude Web Search

Anthropic has integrated a web search function into the Claude model, enabling it to retrieve up-to-date information from the internet. Each response includes inline citations, making it easier to verify sources. This enhancement can improve Claude's accuracy and usefulness in research tasks.

3. OpenAI Audio Models

OpenAI has introduced three new audio models: gpt-4o-mini-tts (text-to-speech) and gpt-4o-transcribe and gpt-4o-mini-transcribe (speech-to-text), which outperform the Whisper model in terms of performance. They are available via the API, which could streamline voice applications such as assistants and transcriptions.

4. Krea AI Video Training

Krea AI has launched a video training feature for the Wan 2.1 model, allowing users to train the AI with their own videos and images to generate personalized video content. This opens up new possibilities for creators, enabling the customization of styles and movements.

5. Mind Maps in NotebookLM

Google's NotebookLM has added a mind map feature, powered by Gemini Pro, helping to visualize and organize complex information. This strengthens NotebookLM's role as an AI-powered research assistant, particularly useful in data analysis.

6. Hunyuan 3D Generation AI

Tencent has updated its Hunyuan 3D generation model, introducing two versions: 3D 2.0 MV (Multi-View Generation) and 3D 2.0 Mini for faster generation. This could accelerate the development of 3D applications such as games and visualizations.

7. New Stability AI Virtual Camera

Stability AI has presented Stable Virtual Camera, a tool that transforms 2D images into 3D videos with controlled camera movements, such as 360° rotations, spirals, and zooms. This could revolutionize video content creation, especially in marketing and entertainment.

8. Gemini Canvas and Audio Overview

Google's Gemini now offers Canvas for real-time writing, editing, and previewing code or documents, and Audio Overview generating summaries in the form of AI-hosted podcasts:

• Audio Overviews turn documents into podcasts

• Shareable deployments on subdomains

• Clean, modern UI

• Deep Reasoning search.