Tencent released open-source Hy4 preview model

Tencent unveiled Hy4 preview, a 770B-parameter model with a 1M-token context, built for software, office, gaming, and research work. It outperformed GLM 5.3 and Kimi K3 in internal tests and is now open source.

🗞 #ai @testingcatalog

First outputs from GPT-6 "Astra" model from OpenAI

OpenAI appears to be nearing Astra’s release as internal testing expands around “mozaik-alpha-fdm.” Early zero-shot samples suggest strong coding and visual creation, though deployment may depend on safety and government review.

🗞 #openai @testingcatalog

ASTRA 🔥: First outputs of the upcoming unreleased model from OpenAI have arrived, and these results are stunning!

> Internally named "mozaik-alpha-fdm", rumored to be a checkpoint of the "Astra" model, mentioned by OpenAI earlier as a substantial step-forward improvement.

Voxel Castle from the zAI Discord 👀

Google is testing different models on Google AI Studio Build internally via a side to side comparison.

> Generate two responses to your next prompt and pick the one you prefer.

Flash vs Flash 👀

Google prepares Interactive Reports for Gemini Notebook

Google is preparing Gemini Notebook Interactive Reports, a feature for creating research-focused reports with embedded Studio content. Release messaging suggests an upcoming rollout soon.

🗞 #google @testingcatalog

GOOGLE 🔥: Interactive Reports are being prepared for the upcoming release in Gemini Notebook.

> New: You can now create interactive reports.
> An interactive report with embedded Studio content.

Users can customize these reports by providing a custom prompt and selecting a language.

Based on the suggested prompt, this feature lets users generate interactive user research reports, and complex visualizations.

* Not available to the public yet.

Anthropic is working on an option to run commands in a local sandbox for Claude Code desktop.

> Runs commands from Claude Code in an isolated sandbox. Takes effect for new sessions.

A strict sandbox mode is available too.

> Blocks commands that can’t run inside the sandbox (ssh, for example) instead of running them outside it or asking you. Applies to every Claude Code session on this computer, including the terminal, unless a project’s own settings override it.

Perplexity prepares Hybrid mode for Computer on Mac

Perplexity is preparing Hybrid Mode for Mac, combining cloud orchestration with local inference for lighter tasks, lower credit use, and added privacy. A Privacy Gate may screen sensitive data before any cloud transfer.

🗞 #perplexity @testingcatalog

PERPLEXITY 🔥: A new Hybrid mode is coming soon on Mac, letting local models handle some subtasks.

> Gemma 4 (for 16 GB machines), Qwen3.6, and Perplexity's own model (For 32 GB machines) are currently supported.

> "In Hybrid mode, Computer orchestrates in the cloud and delegates
suitable subtasks to a model on your Mac."

> "Local models run on your Mac's own memory and processor — the more capable your machine, the larger the models it can run."

> "Privacy Gate is a small model that runs entirely on your Mac. It inspects anything headed to the cloud for personal or sensitive information — and if it finds any, you decide how to handle it before anything is sent."

That would be huge 👀

Google develops AI Rooms for Gemini Enterprise

Google is testing Gemini Enterprise Rooms, a prototype shared workspace that combines team context, files, goals, playbooks, and Gemini in one place.

🗞 #google @testingcatalog

GOOGLE 🔥: A new "Rooms" concept is being prototyped on Gemini Enterprise.

Rooms are based on Projects and allow users to set Goals and Playbooks for Gemini to follow.

Gemini will act as an expert in the provided knowledge base to "keep the team focused."

It's like a war room for AI and humans 👀