Exclusive: Microsoft tests new MAI Realtime voice model
SCOOP 🔥: Exclusive early preview of “MAI-Realtime”, an upcoming bidirectional voice model from Microsoft!
The model feels much more natural than MAI Voice 2 and can operate web search, along with many other tools.
🗞 #microsoft @testingcatalog
SCOOP 🔥: Exclusive early preview of “MAI-Realtime”, an upcoming bidirectional voice model from Microsoft!
The model feels much more natural than MAI Voice 2 and can operate web search, along with many other tools.
🗞 #microsoft @testingcatalog
TestingCatalog AI News
Exclusive: Microsoft tests new MAI Realtime voice model
Two voices are present so far - Victoria and Grant, both noticeably more natural than what Copilot’s voice mode currently delivers.
🚨 AI News | TestingCatalog
Exclusive: Microsoft tests new MAI Realtime voice model SCOOP 🔥: Exclusive early preview of “MAI-Realtime”, an upcoming bidirectional voice model from Microsoft! The model feels much more natural than MAI Voice 2 and can operate web search, along with many…
Media is too big
VIEW IN TELEGRAM
> MAI Realtime is currently being tested on MAI Playground via limited access.
> Two voices are available: Victoria and Grant. It supports automatic language detection and switching to another language mid-conversation.
> Two voices are available: Victoria and Grant. It supports automatic language detection and switching to another language mid-conversation.
👍3
🚨 AI News | TestingCatalog
> MAI Realtime is currently being tested on MAI Playground via limited access. > Two voices are available: Victoria and Grant. It supports automatic language detection and switching to another language mid-conversation.
Media is too big
VIEW IN TELEGRAM
> It supports English, German, Spanish, French, Italian, Portuguese, Japanese, Korean, Chinese, Dutch, Hindi, Indonesian, Arabic, Russian, Turkish, Vietnamese, and Thai.
👍5❤4
Google is working on Plugins for Gemini Enterprise
Google is preparing Plugins and a Notifications section for Gemini Enterprise, pointing to packaged workplace workflows built from Skills and Connectors.
🗞 #google @testingcatalog
Google is preparing Plugins and a Notifications section for Gemini Enterprise, pointing to packaged workplace workflows built from Skills and Connectors.
🗞 #google @testingcatalog
TestingCatalog AI News
Google is working on Plugins for Gemini Enterprise
Google is developing Plugins and a dedicated Notifications area for Gemini Enterprise, aiming to support workplace workflows with packaged tools.
👍4❤3
SpaceXAI is working on Orbs for its voices on Grok for iOS.
Every voice has its own color and the design is quite close to how it looks on xAI Console.
Custom voice agents soon? 👀
Every voice has its own color and the design is quite close to how it looks on xAI Console.
Custom voice agents soon? 👀
❤3👀2
Media is too big
VIEW IN TELEGRAM
The upgraded Gemini Notebook experience is now available to all Pro and Ultra users! Gemini can now generate various file types and conduct research directly from the chat.
Now we need apps 👀
Now we need apps 👀
❤7🆒2🤔1👀1
OpenAI & Anthropic 🤖
AI Security Institute published a report clarifying instances in which AI agents from OpenAI and Anthropic engaged in malicious activity during another security evaluation.
What these AI agents did so far 👀
1. An attempted supply-chain attack on real open-source software.
2. Attempts to deceive and target real people (social engineering).
3. Attempts to plant and prompt-inject malicious code.
4. Collaboration between independent agents being assessed simultaneously.
> Almost all of this behaviour (17 actions) came from a single model, Anthropic's Mythos 5, with 2 actions involving OpenAI's GPT-5.6-Sol with cyber classifiers disabled.
OpenAI and another lab 💀
AI Security Institute published a report clarifying instances in which AI agents from OpenAI and Anthropic engaged in malicious activity during another security evaluation.
What these AI agents did so far 👀
1. An attempted supply-chain attack on real open-source software.
2. Attempts to deceive and target real people (social engineering).
3. Attempts to plant and prompt-inject malicious code.
4. Collaboration between independent agents being assessed simultaneously.
> Almost all of this behaviour (17 actions) came from a single model, Anthropic's Mythos 5, with 2 actions involving OpenAI's GPT-5.6-Sol with cyber classifiers disabled.
OpenAI and another lab 💀
❤2👀2
Alibaba released Qwen3.8-Max with open weights coming soon
ICYMI: Alibaba released Qwen3.8-Max, a 2.4T-parameter model for coding, research, professional work, and multimodal tasks. It is live on QwenCloud now, with open weights for Qwen3.8-Max and Qwen3.8-27B due next week.
🗞 #ai @testingcatalog
ICYMI: Alibaba released Qwen3.8-Max, a 2.4T-parameter model for coding, research, professional work, and multimodal tasks. It is live on QwenCloud now, with open weights for Qwen3.8-Max and Qwen3.8-27B due next week.
🗞 #ai @testingcatalog
TestingCatalog AI News
Alibaba released Qwen3.8-Max with open weights coming soon
Qwen3.8-Max is Qwen’s first Max-scale model slated for open weights, with 2.4 trillion parameters and availability on QwenCloud now.
👍4
OpenAI is working on a new download page for Rosalind Codex, a Codex build configured for GPT-Rosalind.
This change marks a potential expansion beyond "qualified customers".
> GPT-Rosalind is a frontier reasoning model built to support research across biology, drug discovery, and translational medicine.
This change marks a potential expansion beyond "qualified customers".
> GPT-Rosalind is a frontier reasoning model built to support research across biology, drug discovery, and translational medicine.
❤7👍3
SpaceX 🤝 NVIDIA
SpaceX partnered with NVIDIA to design the Starmind AI1 satellite compute payload.
> The AI1 satellite's compute payload will be powered by the NVIDIA Vera Rubin NVL72.
> Starmind will capture solar energy in space to power low-cost, high-performance AI compute on Earth.
SpaceX partnered with NVIDIA to design the Starmind AI1 satellite compute payload.
> The AI1 satellite's compute payload will be powered by the NVIDIA Vera Rubin NVL72.
> Starmind will capture solar energy in space to power low-cost, high-performance AI compute on Earth.
❤6👍4
This media is not supported in your browser
VIEW IN TELEGRAM
BFL released FLUX 3 Video model 👀
> i2v and t2v with native audio, frames, and stuff. Up to 20 seconds and 1080p native.
> A new Draft mode to get a brief sample before spending a full cost on a video gen.
> 2K, 4K, and Open Weights coming soon!
> i2v and t2v with native audio, frames, and stuff. Up to 20 seconds and 1080p native.
> A new Draft mode to get a brief sample before spending a full cost on a video gen.
> 2K, 4K, and Open Weights coming soon!
❤6👍5
🚨 AI News | TestingCatalog
SpaceXAI is working on Orbs for its voices on Grok for iOS. Every voice has its own color and the design is quite close to how it looks on xAI Console. Custom voice agents soon? 👀
SpaceXAI is rolling out loads of new voices for Grok app on iOS along with an updated UI.
Space orbs 👀
Space orbs 👀
❤5👍5
Media is too big
VIEW IN TELEGRAM
BREAKING 🔥: ByteDance launched SeedRealtime, a native audio-visual full-duplex LLM!
> SeedRealtime uses a unified architecture to natively fuse audio, video, and text, enabling real-time interaction over continuous multimodal streams and delivering a brand-new "watch, listen, and speak" experience.
It is now live on the Doubao App for free, enabling voice and video conversations.
Full-duplex Omni 🤯
> SeedRealtime uses a unified architecture to natively fuse audio, video, and text, enabling real-time interaction over continuous multimodal streams and delivering a brand-new "watch, listen, and speak" experience.
It is now live on the Doubao App for free, enabling voice and video conversations.
Full-duplex Omni 🤯
❤7👍4
Mistral releases Shieldstral for multimodal moderation
Mistral released Shieldstral, a 3B Apache 2.0 open-weight multimodal safety classifier for text and images. It uses prompt-defined yes/no policies, runs on a single 16GB GPU, and targets customizable moderation without retraining.
🗞 #mistral @testingcatalog
Mistral released Shieldstral, a 3B Apache 2.0 open-weight multimodal safety classifier for text and images. It uses prompt-defined yes/no policies, runs on a single 16GB GPU, and targets customizable moderation without retraining.
🗞 #mistral @testingcatalog
TestingCatalog AI News
Mistral releases Shieldstral for multimodal moderation
Mistral released Shieldstral, a 3B open-weights safety classifier for customizable text and image moderation, under Apache 2.0 on Hugging Face.
❤4👍4
Cloudflare announces Wallets for AI agent payments
Cloudflare Wallets will let AI agents pay for APIs and web services via stablecoins and x402, with owner-set budgets, allowlists, and caps. The system adds optional agent identity and supports Cloudflare’s push into machine-native commerce.
🗞 #ai @testingcatalog
Cloudflare Wallets will let AI agents pay for APIs and web services via stablecoins and x402, with owner-set budgets, allowlists, and caps. The system adds optional agent identity and supports Cloudflare’s push into machine-native commerce.
🗞 #ai @testingcatalog
TestingCatalog AI News
Cloudflare announces Wallets for AI agent payments
Cloudflare announced programmable Wallets for AI agents, with handles claimable now and stablecoin and x402 funding and payment features planned.
👍4❤3