Mistral Launches 8X Cheaper AI Model

AI is leveling up—faster design, cheaper models, and agents that browse for youThis week’s updates spotlight a major leap in AI performance and usability. Mistral's new Medium 3 model outperforms Llama 4 while slashing costs, making it ideal for enterprise-scale deployment. At Config 2025, Figma introduced powerful AI features like prompt-to-code site generation and vector editing magic—all within your design canvas. And Hugging Face launched a free AI agent that runs a virtual Linux machine through plain language, hinting at the future of browser automation. Here's what's new in the world of AI.

Mistral Medium 3 outperforms Llama 4

Mistral AI has announced Mistral Medium 3, a new language model that delivers state-of-the-art (SOTA) performance at 8X lower cost compared to competitors, while offering radically simplified enterprise deployment options. Positioned as a balance between high performance and affordability, Mistral Medium 3 is optimized for professional use cases such as coding and multimodal understanding, and supports hybrid, on-premises, or in-VPC deployment, as well as custom post-training and integration into enterprise systems. The model achieves over 90% of Claude Sonnet 3.7’s performance on benchmarks but at a much lower price ($0.4 input / $2 output per million tokens), outperforms leading open models like Llama 4 Maverick and enterprise models like Cohere Command A, and can be self-hosted on any cloud with four GPUs or more. It excels particularly in coding and STEM tasks, as confirmed by both internal benchmarks and third-party human evaluations, and is already being used by beta customers in sectors such as finance, energy, and healthcare to enhance customer service, personalize business processes, and analyze complex data. Mistral Medium 3 is available now via Mistral La Plateforme and Amazon Sagemaker, with support for other major cloud platforms coming soon.

Figma Unveils Powerful New AI Features at Config 2025

At Config 2025, Figma launched a suite of innovative AI-powered features designed to streamline the entire design-to-production process. The highlight is Figma Make, an AI prompt-to-code tool that allows users to turn their designs into fully interactive, coded prototypes using simple natural language instructions, making it easier than ever to bridge the gap between design and development. Alongside this, Figma introduced Figma Sites, which lets teams publish responsive, interactive websites directly from Figma with just one click, and Figma Draw, a revamped vector editing suite for greater creative freedom. These AI-driven updates are set to empower designers and product teams to move from idea to production faster, more collaboratively, and with unprecedented creative control-all within the Figma platform.

Hugging Face Launches Free AI Agent for Web Tasks

Hugging Face has released Open Computer Agent, a free, cloud-hosted AI tool that operates a Linux virtual machine through natural language commands, similar to OpenAI’s Operator. Accessible via a web browser, the agent can perform simple tasks like opening Firefox and searching Google Maps, but struggles with more complex requests and often fails to solve CAPTCHAs. Users may experience wait times due to virtual queues. The release aims to showcase the growing capabilities and cost-efficiency of open-source AI models powering agentic workflows, which use vision models capable of interacting with on-screen elements by their coordinates. Although still experimental and somewhat slow, Open Computer Agent highlights the rising interest and investment in AI agents, with 65% of companies experimenting with such technology and the market projected to grow from $7.84 billion in 2025 to $52.62 billion by 2030. Hugging Face’s open-source approach encourages community collaboration to improve AI-driven computer automation and productivity tools.

Hand Picked Video

In this video, we'll look at Mistral's AI Agents.

Top AI Products from this week

• AskTube.Ai - AskTube.Ai makes it easy to talk to YouTube videos and get precise answers. Summarize content, cite sources, receive step-by-step guides, and preview if a video’s worth your time and in your language. Save chats with history & favorites to comeback later.

• Chef by Convex - Chef by Convex is an AI agent that builds real-time full-stack applications. Describe the app you want to build. Chef turns your prompt into a working app with a backend, database, auth, file storage, cron jobs, and a deployed frontend.

• PixelGlow - PixelGlow lets you cut out objects from different images, combine them on a new background, and use AI to blend them seamlessly, making them look naturally captured in a single photo. Instantly create professional mockups and visuals right from your phone.

• Promptify - Promptify is your AI sidekick — accessible from any app or website with a single shortcut (Cmd + I). It’s designed to keep you in flow while helping you write, summarize, brainstorm, and code — without breaking focus or switching tabs.

• Oppla AI - Meet your full‑stack agentic AI product team. Oppla AI deploys a full AI product team—PM, growth, marketing, UX & analytics agents that blaze through discovery‑to‑launch cycles nonstop. Plan, prioritize, build, optimize and ship weeks of work overnight.

• Creation Studio 1.0 - Chat the idea, launch the agent. Build and deploy AI agent systems in minutes just by conversations.

This week in AI

• AI Pathology Models with DINOv2 - Mahmood Lab built open-source pathology foundation models using DINOv2, enabling diverse, data-efficient AI for disease diagnosis, prognosis, and clinical decision support.

• Figma Unveils AI 'Vibe-Coding - Figma Make lets users describe or upload designs to instantly generate working app or website code with AI, available to premium subscribers only.

• Stripe Launches AI & Stablecoin ToolsStripe unveiled a payments foundation model trained on billions of transactions, plus stablecoin financial accounts for businesses in 101 countries, announced May 7, 2025.

• Gemini 2.0 Flash Adds Image Creation - Gemini 2.0 Flash now lets developers generate and edit images via API, with better quality, accurate text, and conversational editing in Google AI Studio and Vertex AI.

• AI Agents Pose New Cybersecurity Risks - Autonomous AI agents can cause data breaches, misuse credentials, and leak sensitive info if not secured. Experts urge dynamic identity management and strict guardrails as deployments rise.