In partnership with

Welcome back!

Google is making image editing faster and more precise with the new version of Gemini 2.0 Flash, allowing users to precisely tweak parts of an image while preserving the rest. It's up to 10x faster than GPT-4o at image editing. Speed and surgical precision are Gemini’s competitive edge over style and image size. Let’s unpack…

Today’s Summary:

• ✂️ Gemini 2.0 improves conversational image editing

• 🧑‍💻 ChatGPT adds PDF report export

• 🩺 OpenAI launches medical AI benchmark

• 🛒 Instacart CEO joins OpenAI

• 🦜 Nvidia transcribes audio instantly

• 🧱 LegoGPT turns text into models

• 🛠️ 2 new tools

TOP STORY

Gemini 2.0 Flash improves conversational image editing

The Summary: Google has released upgraded image generation and editing features in Gemini 2.0 Flash. The new version improves visual quality, reduces safety filtering, and supports conversational editing. It’s still slightly behind ChatGPT in image quality, but it’s faster and better at editing only parts of an existing image, leaving the rest unchanged.

Key details:

• Supports real-time, part-specific image editing with text rendering

• Up to 10x faster than ChatGPT image generation, though visuals are slightly lower in quality

• Performs better on existing image edits, whereas GPT-4o tends to regenerate the whole image every time regardless of instructions

• Available in the Gemini app, Google AI Studio and Vertex AI as gemini-2.0-flash-preview-image-generation

• Not currently available in the EU / UK

Why it matters: Gemini 2.0 Flash image editing is focused on speed and precision rather than style. That makes it a strong candidate for integration into real-world image editing workflows. Multimodal image editing is quickly becoming the next creative interface for manipulating visual media.

Try it in AI Studio

OPENAI

ChatGPT adds PDF export to Deep Research

The Summary: OpenAI just made Deep Research reports downloadable as fully formatted PDFs with citations, images, and tables. This long-missing feature makes ChatGPT far more practical for professionals who need to share research with clients or teams. It’s a clear signal that OpenAI is getting serious about enterprise workflows.

Key details:

• The “Download as PDF” export also works for previously generated Deep Research reports, available now for Plus, Team, and Pro users

• PDF reports retain formatting, clickable citations and images

• The feature mirrors Perplexity which already included PDF export

Why it matters: Most workplaces run on PDFs. By letting users export polished, sourced reports, OpenAI boosts Deep Research from a cool demo into something that begins to fit into consulting decks, client briefings, and compliance workflows. OpenAI is no longer just building AI models, but also practical tools that serve professionals.

FROM OUR PARTNERS

Built for Independents. Powered by AI.

Unlock AI-powered productivity

HoneyBook is how independent businesses attract leads, manage clients, book meetings, sign contracts, and get paid.

Plus, HoneyBook’s AI tools summarize project details, generate email drafts, take meeting notes, predict high-value leads, and more.

Think of HoneyBook as your behind-the-scenes business partner—here to handle the admin work you need to do, so you can focus on the creative work you want to do.

Try HoneyBook for free

MEDICAL AI

OpenAI releases HealthBench to audit Medical AI

The Summary: OpenAI has released HealthBench, a new benchmark for evaluating how AI systems handle complex medical advice. Built with input from 262 doctors across 60 countries, it evaluates 5,000 real-world scenarios using grading criteria written by physicians. The benchmark is open-source and aims to push safety and relevance in AI healthcare. OpenAI o3 model currently leads the benchmark.

Key details:

• Includes 5,000 conversations graded across 48,000 criteria

• o3 model currently leads the benchmark

• AI-assisted drafts can help doctors refine written responses

• Top models still struggle in uncertain situations

Why it matters: HealthBench is a diagnostic stress test for the future of AI in healthcare. Based on the kinds of messy, real-world cases that clinicians face, it reveals where today’s models succeed or fall short. This matters because many are already using AI for symptom checks and emergencies. Most evaluations until now have focused on simple flashcard-style correctness, while HealthBench rewards complex clinical judgment, depth, and nuance.

Full paper

QUICK NEWS

Quick news

• Instacart CEO joins OpenAI as CEO of Applications

• Nvidia Parakeet v2 transcribes 1 hour of audio in 1 second

• LegoGPT generates buildable Lego designs from text

TOOLS

🥇 New tools

• Remio - Local AI note taker & personal knowledge hub for macOS

• Pipedream MCP - Add 2500+ API to your AI assistant for free