🚨 Open-source is catching up, and Kimi K3 boom

Greetings! This week brought quite a surprise: Kimi K3 has launched, and its benchmark results are massive. Community feedback on its performance has also been highly positive, with people already building some genuinely exciting projects using it. Compared with Fable 5, it looks very solid, too.

It was also a big win for the ChatGPT desktop app. It is finally becoming a proper replacement for the web version, with conversations and projects fully synced across desktop, web, and mobile. Let's break everything down!

Newsletter Guide

Here's your guide on how to make the most out of this newsletter:

🤳 Gear up with your mobile phone and desktop devices.
📩 Scan through the newsletter, picking out the apps that pique your interest.
🔥 Keep an eye on the "hot" emoji to identify the hottest user experiences.
👀 Dive deeper into our TestingCatalog posts to discover how to put these apps to the test.
📲 Roll up your sleeves and try them out yourself!

This weekly newsletter was sponsored by Atimic, NoimosAI, Raft, ClickUp, Rokid:

Anthropic

On the Anthropic side, things were relatively quiet this week. The company has been managing community expectations and communicating that Fable 5 will remain exclusive to the Max and Team Premium subscription plans, which is still good news for power users. Anthropic is also extending the higher usage limits, giving users more time to enjoy Fable 5 while they can.

As for future releases, we received an update about Conway, the managed agent Anthropic had been testing internally with employees and a friends-and-family cohort. It appears that the program is now winding down. Conway will be discontinued on July 24, and users have been advised to export their data.

This could mean one of two things. Anthropic may be discontinuing the Conway project entirely, which would be disappointing given its potential. Alternatively, the company may simply be ending the internal testing program ahead of a broader preview release.

And, of course, we are still expecting Claude Opus 5. Traces of the model have been spotted in several places, suggesting that Anthropic is actively working on it. We may even hear something official in July, so we will have to wait and see.

Google

Google plans to upgrade Gemini Enterprise connectors - Google Gemini Enterprise is rolling out a glowing prompt bar and updating Workspace connectors, preparing support for the Gemini Spark agent.

Google preparing Gemini Live and Skills for web rollout - Google is testing Gemini Live on desktop and web, plus new chat-level skills that may soon let users upload or create reusable tasks.

Moonshot AI

Early look at Kimi K3 generations from Moonshot AI on Arena - Moonshot AI’s Kimi K3 launch seems imminent, with leaks and teasers hinting at a groundbreaking multi-trillion-parameter MoE model.

OpenAI

In the OpenAI world, things are still settling after the release of GPT-5.6. At this point, GPT-6 is the next major model upgrade to watch.

Much of OpenAI’s current effort appears to be focused on the recently released ChatGPT desktop app. It has finally become a fully featured application, with projects, conversations, and other functionality available just as they are on the web version.

OpenAI also appears to be reviewing feedback on the recently updated GPT Live 1 and the new voice mode. This is a positive development, as the experience remains quite limited in its current form. Conversations with the model feel natural, and it is remarkably easy to talk to. However, the experience is not yet particularly actionable. Users are still limited to voice-to-voice interactions because the model cannot reliably capture notes and place them into the chat for later copying.

Eventually, we may be able to control remote agents through voice. That capability could arrive alongside a more polished real-time voice feature for Codex, although the timing and implementation remain unclear.

OpenAI unveils GPT-Red to boost AI safety with internal red-teaming - OpenAI introduces GPT‑Red, a specialized internal system for uncovering vulnerabilities in AI models using red-teaming and self-play learning.

OpenAI prepares Codex Micro keypad to control AI Agents - OpenAI launches the Codex Micro keypad, bringing real-time RGB agent status and customizable controls to developers managing multiple AI agents.

Perplexity

Perplexity launches SPACE runtime for AI agent tasks - Perplexity unveils SPACE, a new sandboxed runtime that powers all Computer sessions and enables persistent AI agent operations at scale.

Sesame

Sesame is adding new Calendar and Email connectors for its iOS agent app.

SpaceXAI

Telegram adds rich text editor, communities, more GIFs - Telegram’s latest update adds a rich text editor for long posts, new Communities, private bot replies and faster GIF search with 350m animations.

Thinking Machines

Thinking Machines debuts open-weight Inkling AI model - Thinking Machines Lab launches Inkling, its open-weights Mixture-of-Experts AI model, offering broad customization and multi-modal support.

Happy testing! 🤖

P.S.

If you use Telegram, don't forget to join us there too! It is the fastest place to learn about new AI scoops 🗞️

P.S.S.

If you want to promote your AI tool in this newsletter, please check our Stan Store for Advertisers.