# AI 24 Updates · 2026-11-20-0000

Published: November 20, 2026 · 00:00 CET

## In Brief

- OpenAI GPT-Live-1.1 — voice model now in the API, with a cheaper standby mode
- Gemini app for Mac — computer use rolls out to paid subscribers in the U.S.
- Runway Praxis-1 — 27B video-action model for robot control, with technical report
- Amazon Polly — voice cloning now generally available as a gated release
- ChatGPT Health — expands to desktop, Android, dots and Sites
- AWS Marketplace — AI agents can pay per request via x402 through AgentCore payments
- Nous Research Hermes Enterprise — Hermes Agent comes to Microsoft Teams, Microsoft Store and Azure
- NVIDIA and Microsoft — Vera Rubin NVL72 VMs on Azure and NVIDIA open models in Microsoft Foundry
- Amazon Connect Resolutions — agentic solution for high-volume operational work is generally available
- AWS Marketplace agent mode — now assesses seller security, buys and deploys solutions in one conversation

---

## 1. OpenAI GPT-Live-1.1 — voice model now in the API, with a cheaper standby mode

Company: OpenAI

OpenAI released GPT-Live-1.1 (gpt-live-1.1), its full-duplex voice model for multilingual conversations, in the API.

**Facts:**
- GPT-Live-1.1 improves pronunciation, accent consistency and transcription in multilingual voice conversations, according to the OpenAI API changelog.
- Developers use gpt-live-1.1 in an existing Live session configuration (v1/live/sessions) and describe in plain language when to delegate to a backend model.
- A new standby option for GPT-Live-1 and GPT-Live-1.1 WebRTC sessions pauses inference after an idle timeout of 5–60 seconds and resumes when someone speaks.
- GPT-Live bills voice sessions per second: $0.05 per minute at the active rate and $0.02 per minute at the standby rate.
- OpenAI says GPT-Live-1.1 has powered voice in ChatGPT Plus and Pro since October 15.

**Sources:**
- [OpenAI API changelog](https://developers.openai.com/api/docs/changelog) (official)
- [GPT-Live-1.1 model page](https://developers.openai.com/api/docs/models/gpt-live-1.1) (official)
- [OpenAI Developers on X](https://x.com/OpenAIDevs/status/2123506733524136446) (official)
- [OpenAI Developers on X: standby](https://x.com/OpenAIDevs/status/2123506763144245744) (official)

## 2. Gemini app for Mac — computer use rolls out to paid subscribers in the U.S.

Company: Google

Google added computer use to the Gemini app for Mac, letting Gemini work across folders, apps and the web in a background window.

**Facts:**
- Gemini can hand off tasks across the user's folders, apps and the web while the user's own screen and cursor stay free.
- Gemini clicks, scrolls and types in a live background window that the user can watch.
- Google's demo shows Gemini turning a quick sketch into a rigged 3D model in Blender.
- Computer use is rolling out in the U.S., in English, to Google AI Plus, Pro and Ultra subscribers in the Gemini app for Mac.

**Sources:**
- [Google blog: computer use](https://goo.gle/GeminiComputerUse) (official)
- [Gemini app on X](https://x.com/GeminiApp/status/2123513377268109331) (official)

## 3. Runway Praxis-1 — 27B video-action model for robot control, with technical report

Company: Runway

Runway published a technical report on Praxis-1, a model that uses video pretraining to control robots.

**Facts:**
- Praxis-1 is a 27B-parameter mixture-of-experts diffusion transformer that predicts video and actions together.
- On the same station and tasks, Runway says Praxis-1 finished 15.6x faster than GPT-6 Astra driving the arm, at 51x lower cost per completed task.
- With under 30 minutes of teleoperated demos, Praxis-1 succeeded in all 15 simulated trials on Noble Machines' robot, according to Runway.
- The technical report details the model's architecture, training stages and results.

**Sources:**
- [Runway research: Praxis-1](https://runway.com/research/praxis-1-video-pretraining-to-robot-control) (official)
- [Runway on X](https://x.com/runwayml/status/2123504522140266636) (official)

## 4. Amazon Polly — voice cloning now generally available as a gated release

Company: Amazon

AWS made voice cloning generally available in Amazon Polly as a gated release.

**Facts:**
- Customers can create a cloned voice from a few minutes of recordings of a consenting speaker.
- Cloned voices run on the Amazon Polly Generative engine.
- Voice cloning launches in seven languages: English, Spanish, German, French, Italian, Portuguese and Hindi; each cloned voice speaks the language it was recorded in.
- Cloned voices are encrypted and access-controlled; AWS says Polly does not keep the recordings after the voice is created and does not use them to train Polly models.
- It is available in 11 AWS Regions, including US East (N. Virginia), US West (Oregon), Europe (Frankfurt) and Asia Pacific (Tokyo).

**Sources:**
- [AWS What's New](https://aws.amazon.com/about-aws/whats-new/2026/11/amazon-polly-now-supports-voice-cloning/) (official)

## 5. ChatGPT Health — expands to desktop, Android, dots and Sites

Company: OpenAI

OpenAI made ChatGPT Health available in more places, including the desktop app, dots, Sites and Android.

**Facts:**
- On desktop, users can analyze sleep and workout patterns from their connected health information.
- Dots can give proactive suggestions, such as another time for a workout when a meeting gets in the way.
- Users can build a Site to track training, such as runs and progress toward a marathon plan.
- On Android, Health connects sleep, activity, workouts and more through Android Health Connect.
- Health is available in the U.S. on desktop and Android for Free, Go, Plus and Pro users; Sites require Plus or Pro, and dots require Pro.

**Sources:**
- [ChatGPT on X](https://x.com/ChatGPT/status/2123545961381216325) (official)

## 6. AWS Marketplace — AI agents can pay per request via x402 through AgentCore payments

Company: Amazon

AWS Marketplace now supports agentic payments over the x402 protocol through AgentCore payments, part of Amazon Bedrock AgentCore.

**Facts:**
- AI agents can discover and buy select products and pay per request in US dollars, with no per-seller sign-up or API key setup.
- Charges settle on the customer's existing AWS bill, with no crypto wallet required.
- Agents find listings through the AWS Marketplace connector in AgentCore Gateway after the customer enables the AWS Marketplace payment connector.
- Customers can set spending limits, approved sellers and guardrails; transactions appear in AWS CloudTrail and AWS Cost Explorer.
- Sellers can charge agents per request by configuring agentic payments and integrating the AWS Marketplace x402 SDK.

**Sources:**
- [AWS What's New](https://aws.amazon.com/about-aws/whats-new/2026/11/marketplace-agentic-payments-agentcore/) (official)

## 7. Nous Research Hermes Enterprise — Hermes Agent comes to Microsoft Teams, Microsoft Store and Azure

Company: Nous Research

Nous Research partnered with Microsoft to launch Hermes Enterprise, bringing Hermes Agent to Microsoft Teams, the Microsoft Store and Azure.

**Facts:**
- In Teams, employees hand Hermes work such as financial analysis, document drafts or inbox triage and review the result in the same chat.
- Agent workloads run in isolated Azure Container Apps sandboxes.
- Hermes saves a good way of doing a recurring task as a skill; with Collective Wisdom, the skill can be shared with the team after review.
- Companies with strict data residency or procurement rules can run Hermes Enterprise in their own Azure tenant.
- Nous Research CTO Jeffrey Quesnelle presents the sandboxes at Microsoft Ignite in session BRK448.

**Sources:**
- [Nous Research on X](https://x.com/NousResearch/status/2123535954946257096) (official)

## 8. NVIDIA and Microsoft — Vera Rubin NVL72 VMs on Azure and NVIDIA open models in Microsoft Foundry

Company: NVIDIA

At Microsoft Ignite, NVIDIA and Microsoft expanded their collaboration from Azure infrastructure to Microsoft Foundry and Windows PCs.

**Facts:**
- NVIDIA Vera Rubin NVL72 systems are running in Azure data centers, and the companies are expanding access to Azure virtual machines powered by them (Azure's Vera Rubin preview).
- The companies are working to extend confidential computing to bare-metal and rack-scale AI infrastructure.
- NVIDIA open models and customization capabilities are being made available through Microsoft Foundry; Foundry Managed Compute adds support for deploying custom model weights.
- Microsoft uses NVIDIA Dynamo inference software for post-training and NVIDIA RTX PRO 6000 Blackwell GPUs for Bing search models.
- NVIDIA says EY adapted Nemotron open models with Programmed Training in Foundry, improving accuracy on its evaluation dataset by 55% and lowering cost by 15x versus proprietary frontier models.

**Sources:**
- [NVIDIA blog](https://blogs.nvidia.com/blog/microsoft-ignite-accelerated-computing-azure-windows-pcs/) (official)

## 9. Amazon Connect Resolutions — agentic solution for high-volume operational work is generally available

Company: Amazon

Amazon Connect Customer made Resolutions generally available, an agentic operations solution that resolves high-volume work end-to-end.

**Facts:**
- Resolutions targets work such as refund requests, transaction disputes, fraud investigations and insurance claims.
- Managers describe a process in natural language or upload standard operating procedures, and Resolutions generates a Work plan in minutes, asking clarifying questions about exceptions, decision rules and approval thresholds.
- AI agents gather information, apply policy, act across business applications and ask customers for missing details for each Resolutions Case.
- Each step can run on prebuilt AI agents, agents configured in Amazon Connect Customer or the customer's own, while human specialists handle approvals and exceptions.

**Sources:**
- [AWS What's New](https://aws.amazon.com/about-aws/whats-new/2026/11/amazon-connect-resolutions/) (official)

## 10. AWS Marketplace agent mode — now assesses seller security, buys and deploys solutions in one conversation

Company: Amazon

AWS expanded AWS Marketplace agent mode, its conversational AI experience for procuring cloud solutions, to cover assessment, purchase and deployment.

**Facts:**
- Users describe a use case in natural language to get recommended software, services, AI agents and tools, and multi-product solutions.
- Agent mode can request documents such as SOC 2 reports and give a criterion-by-criterion assessment against the CAIQ cloud security questionnaire.
- It explains pricing and terms and completes the purchase after the user reviews the terms and confirms.
- For supported products, it reviews the AWS environment, proposes a deployment plan with an estimated monthly cost and deploys after approval.
- Agent mode is available in all AWS Regions where AWS Marketplace is available.

**Sources:**
- [AWS What's New](https://aws.amazon.com/about-aws/whats-new/2026/11/aws-marketplace/aws-marketplace-agent-mode-expansion) (official)
