AI Native Weekly Newsletter: 13 June 2025
Contents
- ChatGPT Launches o3-pro: The Most Reliable AI for Complex Tasks
- UK’s Gemini-powered planning tool cuts 2-hour tasks to 40 seconds
- Meta’s V-JEPA 2: AI World Model Achieves 65-80% Robot Success
- ChatGPT’s Advanced Voice Mode Gets Major Upgrade
- Mistral AI Launches Magistral: Transparent Reasoning Model
- ByteDance Unveils Major AI Upgrades: Doubao 1.6 Model and Leading Video Generation
ChatGPT Launches o3-pro: The Most Reliable AI for Complex Tasks
OpenAI’s newest model, o3-pro, is now available for Pro users in ChatGPT and via API. This enhanced version excels at math, science, and coding, consistently outperforming other models in expert evaluations. It prioritizes reliability over speed, making it ideal for challenging questions where accuracy matters most. While it includes web search, file analysis, and visual reasoning capabilities, note that image generation and Canvas features aren’t currently supported.
UK’s Gemini-powered planning tool cuts 2-hour tasks to 40 seconds
The UK’s Extract system, built with Gemini AI, transforms blurry maps and handwritten planning documents into digital data in just 40 seconds instead of 2 hours. Currently tested at 4 councils including Westminster and Hillingdon, it will roll out to all English councils by Spring 2026 to accelerate processing of 350,000 annual planning applications.
Meta’s V-JEPA 2: AI World Model Achieves 65-80% Robot Success
Meta introduces V-JEPA 2, a 1.2B-parameter world model that enables zero-shot robot planning in new environments with 65-80% success rates for pick-and-place tasks. Built on Joint Embedding Predictive Architecture, it advances from the original V-JEPA by adding action-conditioned training with just 62 hours of robot data. The model achieves state-of-the-art performance on Epic-Kitchens-100 action anticipation and video QA benchmarks. Meta also releases three new physical reasoning benchmarks where humans score 85-95% but AI models struggle near chance levels.
ChatGPT’s Advanced Voice Mode Gets Major Upgrade
ChatGPT enhances Advanced Voice Mode for paid subscribers with more natural intonation, realistic cadences, and improved expressiveness. The update adds seamless language translation capabilities that maintain context throughout conversations. While available across all platforms and markets, users may occasionally experience minor audio quality issues and rare hallucinations resembling ads or background sounds—limitations the team is actively addressing.
Mistral AI Launches Magistral: Transparent Reasoning Model
Mistral AI unveils Magistral, their first reasoning model available in two versions: open-source Small (24B parameters) and enterprise Medium. Magistral excels in transparent, multilingual reasoning with Medium scoring 73.6% on AIME2024 (90% with majority voting). Features 10x faster reasoning via Le Chat, supports 8+ languages, and works across domains from legal/finance compliance to creative writing. Small version enables community development while Medium targets enterprise applications requiring traceable decision-making processes.
ByteDance Unveils Major AI Upgrades: Doubao 1.6 Model and Leading Video Generation
ByteDance’s Doubao large language model has been upgraded to version 1.6, achieving significant improvements in reasoning and mathematical capabilities with an 81.5 score on the GPQA Diamond benchmark. The company also launched Seedance 1.0 Pro, a video generation model that ranks first internationally and supports high-quality 1080p video production, alongside a new voice podcast model for instant conversational content generation.