🛒 ChatGPT partners with Instacart for grocery shopping 🛒
Users can now browse products, build carts, and check out without leaving ChatGPT

OpenAI expanded its long-standing relationship with Instacart by integrating grocery shopping directly into ChatGPT. Users can now browse products, build their shopping cart, and complete checkout without switching apps. The feature represents a significant step toward making AI assistants more practical for daily life tasks. This integration builds on previous partnerships between the two companies and positions ChatGPT as a comprehensive lifestyle assistant.

💻 Mistral launches Devstral 2 coding models 💻
New coding-focused models achieve 72% on SWE-bench with terminal automation

Mistral released Devstral 2 with 123B parameters, achieving 72.2% on SWE-bench Verified and ranking among the best open-weight coding models despite its smaller size. The companion 24B Devstral Small 2 scores 68% while running on consumer hardware. Mistral also launched Vibe CLI, an open-source terminal agent that orchestrates multi-file changes across entire codebases, enabling automated code refactoring and development workflows.

🥑 Meta plans "Avocado" model launch 🥑
Company may break open-source tradition with proprietary frontier model

Meta is reportedly planning to launch a new frontier AI model codenamed "Avocado" in Q1 2026, marking a potential shift from its open-source strategy. Sources suggest this could be a proprietary model, unlike previous Llama releases that were made freely available. The development signals Meta's evolving approach to AI competition as the company seeks to maintain its position against OpenAI and Google's latest offerings.

INTERESTING TO KNOW

🔬 Microsoft AI maps cancer tumors 🔬

Microsoft open-sourced GigaTIME, an AI model that transforms basic $10 tissue slides into detailed cancer analysis previously requiring expensive lab work and days of processing. The system learned from 40 million cell samples and was tested on over 14,000 cancer patients, creating a virtual library of 300,000 detailed tumor images across 24 cancer types.

🇨🇳 GLM-4.6V multimodal models released 🇨🇳

Zhipu AI has open-sourced GLM-4.6V multimodal models in two variants: a 106B foundation model for cloud deployment and a 9B lightweight version for local use. GLM-4.6V scales its context window to 128k tokens and achieves state-of-the-art performance in visual understanding among similar-sized models with native Function Calling capabilities.