OpenAI Unveils o3-pro🤯😱

The AI landscape just hit a new gear. OpenAI has rolled out o3-pro, its most advanced reasoning model yet—engineered to handle science, mathematics, business strategy, and complex code with unmatched clarity and step-by-step logic. Replacing the o1-pro for Pro and Team users, and soon reaching Enterprise and Edu tiers, o3-pro dominates benchmarks, outperforming Google’s Gemini 2.5 Pro in math and Anthropic’s Claude 4 Opus in science. It’s equipped with tools like web search, Python execution, and file analysis—but sacrifices speed for brainpower, and image generation is still off the table for now.

At the same time, Mistral AI has made its first major move in the reasoning space with Magistral, a model that prioritizes transparency and multilingual logic. Available in both open-source (Small) and enterprise-grade (Medium) versions, Magistral excels in regulated fields like healthcare and finance where traceability matters. It performs well across reasoning and structured tasks, and supports languages like English, French, Arabic, Spanish, and Chinese—bringing logical clarity to global applications.

And for the creatives, Meta just made video editing a lot more fun. Its new generative video feature, now available in Meta AI tools and apps, lets users transform short videos with 50+ AI-driven effects—outfits, lighting, locations, even entire visual styles like comic book scenes or video game looks. No skills required, just a few taps.

From powerful logic engines to open creativity tools, this week marks a turning point in how we reason, build, and express with AI. The future isn’t just smart—it’s transparent, multilingual, and wildly creative.

Most Advanced Reasoning AI Model

OpenAI has launched o3-pro, its most advanced AI reasoning model to date, designed to outperform previous offerings in complex domains such as science, mathematics, coding, business, and writing. O3-pro is an enhanced version of the o3 model, emphasizing step-by-step logical reasoning for greater reliability in challenging problem-solving tasks. It is now available to ChatGPT Pro and Team users—replacing the o1-pro model—with Enterprise and Edu users gaining access the following week, and is also live in OpenAI’s developer API. O3-pro is priced at $20 per million input tokens and $80 per million output tokens, with a million input tokens equating to roughly 750,000 words. According to OpenAI, expert reviewers consistently rated o3-pro higher than o3 across all tested categories, particularly for clarity, comprehensiveness, instruction-following, and accuracy. The model supports advanced tooling, such as web search, file analysis, visual reasoning, Python code execution, and personalized memory-based responses. However, it currently cannot generate images, does not support OpenAI’s Canvas workspace, and temporary chats are disabled while a technical issue is resolved. O3-pro’s responses are generally slower than o1-pro due to its more complex reasoning processes. In benchmark tests, o3-pro surpasses Google’s Gemini 2.5 Pro in math (AIME 2024) and outperforms Anthropic’s Claude 4 Opus in PhD-level science (GPQA Diamond), positioning it as a leading choice for enterprise and advanced academic applications.

Mistral AI Launches Magistral: Transparent, Multilingual Reasoning Model

Mistral AI has introduced Magistral, its first AI reasoning model, available in two versions: the open-source Magistral Small (24 billion parameters) and the more powerful enterprise-focused Magistral Medium. Magistral is engineered for transparent, step-by-step reasoning across a wide range of domains, supporting multiple languages including English, French, Spanish, Arabic, and Simplified Chinese. Unlike traditional language models, Magistral emphasizes traceable logic, making it particularly valuable for regulated industries such as law, finance, healthcare, and government, where understanding and verifying the AI’s reasoning is critical. The model is also well-suited for structured calculations, programmatic logic, decision trees, and creative tasks like storytelling. Magistral Medium demonstrates strong benchmark performance, achieving 73.6% on AIME2024 and 90% with majority voting, while Magistral Small scores 70.7% and 83.3% respectively. With features like rapid response in Le Chat and open-source availability for Magistral Small, Mistral aims to accelerate the adoption and development of transparent, domain-specialized AI reasoning models in both enterprise and community settings.

Meta AI Launches Generative Video Editing

Meta has introduced a new generative AI video editing feature available on the Meta AI app, Meta.AI website, and Edits app, allowing users in the US and over a dozen countries to easily transform short-form videos with more than 50 preset AI prompts. This intuitive tool enables anyone to creatively edit up to 10 seconds of video for free, changing elements like outfits, locations, lighting, and visual styles—such as turning a video into a comic book scene or a video game look—without any prior editing experience. Edited videos can be shared directly to Facebook, Instagram, or the Discover feed, making it simple for users and creators to experiment and engage audiences. Inspired by Meta’s Movie Gen models, this launch marks the first step toward more advanced AI-powered video generation and editing, with future updates planned to support custom text prompts for even greater creative control.

Hand Picked Video

In this video, we'll explore how to build an AI-powered sales outreach agent that automates prospect research and email campaigns.

Top AI Products from this week

• undash - Undash is a simple Chrome plugin that automatically removes long em dashes (—) from AI-generated text when you paste it anywhere: websites, social media, emails, and more. No setup needed. Just install, enable, and paste clean text instantly.

• Merlio - Launch smarter with Merlio — your AI command center for content, images, and idea generation. Powered by GPT, Claude, Gemini & more. 🚀 Upvote if you want sharper images, smarter chats & AI squads working for you.

• Kodosumi - Deploy and scale your AI agents with Kodosumi, the open-source runtime built for developers. Fast, scalable, and fully free to use.

• Bridgely - EmotionSense Pro is a Chrome extension that detects real-time emotions in Google Meet using AI—without sending data to servers. Ideal for UX researchers, recruiters, and educators seeking ethical, local emotion analytics.

• Spydr - The internet wasn't designed for an AI-first world. Information is scattered, and your context rests in siloed apps with no way to carry it over. With Spydr Memory MCP, we bring the first multimodal, interoperable context engine for any AI client.

• Virtuall.PRO - Generate 3D models in minutes using leading AI engines. Compare output, refine results, and collaborate in one powerful 3D workspace. From concept to production ready 3D model —faster than ever.

This week in AI

• Meta Forms Superintelligence Lab - Meta launches a new AI lab, enlisting Scale AI CEO Alexandr Wang and considering a major investment in Scale AI, as it aims to outpace rivals in the AI race.

• Next-Gen TTS - Bland’s new TTS engine uses LLMs to generate lifelike, emotionally rich speech from text, enabling precise style control, voice blending, and sound effect integration like never before.

• Copilot Zero-Click Flaw - EchoLeak let hackers steal Microsoft 365 Copilot data via email—no clicks needed. Fixed now, but exposes a major AI agent security risk for all enterprises.

• Huang vs. Amodei on AI Jobs - Nvidia’s Jensen Huang rejects Anthropic CEO’s AI job-loss fears, saying AI will change, not just cut, jobs—and calls for open, transparent AI development.

• Open-Weights Delay - Open-weights model release pushed to later this summer, not June. Team achieved a major breakthrough—delay ensures higher quality and is worth the wait.