OpenAI rolled out GPT-4.5 to all ChatGPT Plus users

OpenAI is rolling out GPT-4.5 to ChatGPT Plus users, with full availability expected within days. The model is more costly and computationally intensive, with mixed benchmark results. It offers deeper knowledge, improved emotional intelligence, and fewer hallucinations.

🗞 #chatgpt

Microsoft working on voice avatars, generative layouts, and agents for Copilot

Microsoft is developing new Copilot features, including a redesigned UI with a prompt bar, chat history, and navigation improvements. Experimental additions include voice-based avatars and AI-generated visuals. Future updates may introduce agentic AI for task automation.

🗞 #microsoftcopilot

Mistral AI expands its AI portfolio with powerful new OCR models

Mistral AI has introduced two OCR models, mistral-ocr-2503 and mistral-ocr-latest, for extracting text from images and documents. They support multiple languages, recognize handwritten text, and preserve formatting. These models cater to industries requiring precise document processing.

🗞 #mistral

Tavus launches three AI models to advance in conversational video

Tavus has introduced three AI models—Phoenix-3, Raven-0, and Sparrow-0—designed to enhance Conversational Video Interfaces. Together, they improve facial animation, visual perception, and conversational flow. These models will be publicly available via API on March 6, 2025.

🗞 #ai

World launches World Chat, secure messaging for verified users

World Network has introduced World Chat, a Mini App enabling secure messaging among verified users with end-to-end encryption. It integrates the World App wallet for cryptocurrency transactions. A color-coded system distinguishes verified users. The launch highlights the expansion of World Network’s Mini App ecosystem, backed by key industry partnerships through World Build. The incubator program has drawn over 150 builders, with selected teams joining a development retreat. World Chat is available on Android and iOS.

🗞 #worldcoin

DuckDuckGo expands AI search as Duck.ai moves out of beta

DuckDuckGo has introduced AI updates prioritizing privacy and user choice. Duck.ai offers anonymized access to chatbots from multiple providers. AI-assisted answers now appear globally and can be customized. Data remains private, with no storage on DuckDuckGo’s servers.

🗞 #duckduckgo

Gemini AI may soon offer personalized responses based on search history

Google is developing the Gemini Personalization model, an experimental AI feature that tailors responses using Google Search history. Users must opt in and enable Web & App Activity. While aiming to refine chatbot relevance, it raises privacy concerns with transparency measures.

🗞 #gemini

Google plans to release new Gemini models on March 12

Google may launch new Gemini models on March 12, based on recent code findings. Potential releases include Flash 2.0 Thinking models and a personalization feature. Google is also working on screen-sharing and real-time video analysis. Delays remain possible.

🗞 #gemini

Google AI Studio may get support for Gemini Embedding and Imagen 3 models

Google AI Studio may introduce Gemini Embedding and Imagen 3 models. Gemini Embedding supports up to 8,000 tokens, outputs 3,072-dimensional vectors, and handles 100+ languages. Imagen 3 offers high-quality image generation with improved detail and watermarking.

🗞 #aistudio

OpenAI released new tools and APIs for AI agent development

OpenAI has introduced new tools and APIs to support AI agent development, including the Responses API and an open-source Agents SDK. These tools enable autonomous task execution using built-in web search, file search, and automation features while improving orchestration and debugging.

🗞 #chatgpt

Gemma 3 sets new benchmarks for open compact models with top score on LMarena

Google has launched Gemma 3, its most advanced open-model with multimodal capabilities, extended context windows, and improved multilingual support. Available in four sizes, it integrates a vision encoder, supports 128k token contexts, and was trained with extensive optimization techniques.

🗞 #aistudio

Google rolls out multimodal image generation for Gemini 2.0 Flash in AI Studio

Google has introduced multimodal image generation and editing in Gemini 2.0 Flash via AI Studio, allowing users to create and modify images without full regeneration. A new "Output Format" toggle enables text-only or text-and-image responses. All images have SynthID watermarks.

🗞 #aistudio

Deep Search may be coming to NotebookLM in future updates

Google is developing new features for NotebookLM, including support for multiple languages and a "Discover Sources" tool for deeper research. These updates, still in development, could expand its research capabilities and align with Google's broader AI strategy.

🗞 #notebooklm

OpenAI works on collaborative tools, referral programs, and native image generation

OpenAI is introducing collaborative workspaces for team users, a referral program targeting students, and native image output for GPT-4o. These updates aim to expand ChatGPT’s functionality for both individuals and organizations, aligning with industry trends and competitive pressures.

🗞 #chatgpt

Google Gemini gets major “Thinking” upgrades and free Deep Research

Google has updated Gemini with a new experimental model, expanding file uploads and a larger context window for advanced users. Deep Research now supports multi-page reports in 45+ languages. Personalization and integration with Google apps have been introduced, while Gems are now free for all users.

🗞 #gemini

Anthropic finalizing Harmony, an AI agent to operate with local files

Anthropic is developing Harmony, a feature enabling Claude to analyze, edit, and index local files. It may serve as an AI coding assistant by identifying vulnerabilities. Another tool, Compass, could facilitate deep research. Web search expansion is also in progress.

🗞 #claude

Baidu unveils ERNIE 4.5 and ERNIE X1, challenging DeepSeek with lower prices

Baidu has launched ERNIE 4.5 and ERNIE X1 AI models ahead of schedule, offering free access to individuals. ERNIE 4.5 is a multimodal foundation model, while ERNIE X1 excels in deep-thinking reasoning. Both models are set for enterprise integration via Baidu AI Cloud.

🗞 #ai

Google prepares Canvas and Veo2 integration for Gemini

Google is developing two new Gemini features: Canvas, enabling document and code file creation, and video generation, likely powered by Veo2. Canvas will first be available on Gemini 2.0 Flash. A video generation placeholder suggests a launch within weeks.

🗞 #gemini

Command A: Cohere’s latest AI model challenges GPT-4o with enterprise focus

Cohere has introduced Command A, a high-performance generative AI model optimized for enterprise tasks with reduced computational demands. Positioned alongside top models like GPT-4o, it supports various industries while integrating with Cohere’s AI ecosystem, prioritizing security and efficiency.

🗞 #ai

Mistral Small 3: A 24B open-source AI model optimized for speed

Mistral AI has introduced Mistral Small 3, a 24B parameter open-source language model focused on low latency and efficiency. It rivals larger models while running over three times faster. With an 81% MMLU score, it supports various AI tasks and is optimized for local deployment. Released under Apache 2.0, it offers pre-trained and fine-tuned checkpoints for flexible use. Target applications include virtual assistants, automated workflows, and industry-specific fine-tuning. Available on multiple platforms, Mistral AI reaffirms its commitment to open-source by phasing out MRL-licensed offerings.

🗞 #mistral