Sunday Rundown #144: Image Upgrades & Pogo Skates

Live & Learn #5: Testing five AI logo makers.

Jun 07, 2026

Happy Sunday, friends!

Welcome back to the weekly AI news roundup.

In case you missed it, here’s this week’s Thursday deep dive:

If you’re consistently missing out on my emails, remember to check your “Promotions” tab and mark whytryai@substack.com as a “Safe Sender.”

👩‍💻 AI releases

Google open-sourced Gemma 4 12B, a multimodal model that processes text, images, and audio natively in a single pass and can run on a home laptop.
H Company open-sourced Holo3.1, a family of AI agents that can click, type, and navigate apps on the web, desktop, and mobile while running on your own device.
Ideogram launched Ideogram 4.0, an image model with the most reliable text rendering of any open-weight release, free for non-commercial use.
JetBrains open-sourced Mellum2, a fast coding model that can act as a specialist in bigger multimodel pipelines to handle routing, summarization, and refactoring.
Krea released Krea 2 Turbo, a faster version of Krea 2 that can create high-quality images in about two seconds. (Try it for free.)
Microsoft rolled out its first in-house coding model, MAI-Code-1-Flash, directly inside GitHub Copilot, free and optimized for fast, lightweight coding tasks.
MiniMax open-sourced M3, a natively multimodal model with 1M context, frontier performance, and coding skills designed for long-running agent tasks.
Nous Research launched Hermes Desktop, a native app for installing, configuring, and chatting with Hermes Agent without using the terminal.
OpenAI news:
1. ChatGPT can now send emails directly from a conversation, so you can draft, edit, and fire off a message without leaving the chat.
2. Codex can now adapt to your specific job role thanks to plugin bundles for marketing, sales, support, and more with relevant apps and skills.
3. Dreaming lets ChatGPT process your past chats in the background to build a persistent understanding of your preferences. (Paid US accounts only for now.)
Reve launched Reve 2.0, a 4K image generator that lets you control exactly where elements appear using visual layouts rather than just text prompts. (Try for free.)
Runway brought its Aleph 2.0 model to the API, letting developers make precise video edits while preserving the rest of the clip.
TwelveLabs launched Rodeo, an AI video copilot that lets you find and assemble footage using plain language without manual scrubbing.
xAI released Grok Imagine 1.5 Preview, which lets you turn still images into cinematic 720p videos with natural-language prompts.

🔬 AI research

GitHub announced a Copilot Desktop App in technical preview, letting developers manage coding agents, create secure sandboxes, and review changes.
Microsoft introduced Scout, an always-on Microsoft 365 agent that schedules meetings, flags stalled decisions, and learns your priorities over time.
Perplexity teased a hybrid inference system for Perplexity Computer that automatically splits AI tasks between your device and the cloud.

📖 AI resources

“EVA-Bench Data 2.0” [BENCHMARK]: a set of 121 tools and 213 real-world scenarios to test AI voice agents across different enterprise domains.

🔀 AI random

Anthropic confidentially filed a draft S-1 for a proposed IPO with the Securities and Exchange Commission. (But let’s hope it gets its house in order soon.)