Genie 4 opens 720p interactive worlds to Gemini API developers
Genie 4 and Polly voice cloning, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
Genie 4
DEPLOYif you prototype interactive experiences
WHAT SHIPPEDGenie 4 is available on the Gemini API through the Interactions API and in Google AI Studio. It generates continuous 720p worlds at 24fps with synchronized native audio from a text prompt, a reference image, or both.
THE CATCHEvery output carries imperceptible SynthID watermarking, and world state consistency is held by Gemini reasoning rather than by your own code.
IF YOU RUNIF YOU RUN robotics or agent training that needs varied simulated environments, you can drive weather, lighting, objects and events live with keyboard controls and text prompts.
SOURCEGoogle, Genie 4 is now available on the Gemini API
Amazon Web Services
Polly voice cloning
WAITif you need a brand voice
WHAT SHIPPEDAmazon Polly voice cloning is generally available as a gated release. You send 1 to 10 recordings of a consenting speaker and Polly builds a cloned voice on the generative engine in seven languages: English, Spanish, German, French, Italian, Portuguese and Hindi.
THE CATCHAccess is enabled for select Amazon Polly customers only, and others must reach out to their AWS account team or apply. Each cloned voice speaks only the language it was recorded in.
IF YOU RUNIF YOU RUN contact center prompts that change with every offer, you can regenerate on-brand audio on demand instead of booking a studio session.
SOURCEAmazon Web Services, Amazon Polly launches voice cloning: A unique voice for your brand
The Deploy Log
Models and APIs
| DEPLOYif you process mortgage or claims packets Textract Insights. Amazon Textract Insights is generally available and takes a single file of up to 3,000 pages, splitting mixed packets and returning a confidence score and page grounding for each value. Amazon Web Services |
| WATCHif you run live customer calls Dialog-RSN-1. PolyAI Dialog-RSN-1 is available on AWS Marketplace and SageMaker JumpStart, deploying as a real-time endpoint that speaks the OpenAI Realtime protocol with connections up to 30 minutes. Amazon Web Services |
| DEPLOYif you run agents on Bedrock AgentCore harness. Amazon Bedrock AgentCore harness now runs on Strands harness with lifecycle hooks and proactive context management, which AWS measures at about 34% lower token cost than other leading harnesses. Amazon Web Services |
| WATCHif you serve models across regions HyperPod Global Inference Gateway. The Global Inference Gateway routes across every registered HyperPod cluster from one endpoint, cutting a team's 20 model-per-cluster endpoints to one and scoring clusters on four weighted signals. Amazon Web Services |
| WATCHif you pay for inference GPUs SageMaker endpoint monitoring. SageMaker AI now monitors vLLM and GPU inference endpoints by default for latency drift, idle GPUs and missed optimizations, flagging cases where the busiest GPU used 12 percent or less of capacity. Amazon Web Services |
| WATCHif you build voice agents GPT-Live 1.1. GPT-Live 1.1 is on AI Gateway as a full-duplex voice model, and an existing GPT-Live integration needs only the new model ID. Vercel |
| WATCHif you classify audio or images Liquid d1.2. Liquid AI's d1.2 and d1.2-ultra are on AI Gateway with audio input and bounding box support, returning structured answers to Boolean, Choice and Score questions. Vercel |
| WATCHif you run models locally Inkling 2 Mini. Inkling 2 Mini is a multimodal Mixture-of-Experts with about 26B total and 3B active parameters, a 1M-token context, and 48 layer visits from 24 physical layers, leading 32 of 48 benchmarks. Hugging Face |
Tools and products
| DEPLOYif you live in the Codex TUI Codex 0.171.0. Codex rust-v0.171.0 adds GPT-6.2 Sol and Astra to model catalogs with Sol as the default, and turns the TUI workspace on by default across 173 commits. OpenAI |
| WATCHif your agents poll pull requests gh event listen. GitHub CLI v2.105.0 adds gh event listen, which waits for pull request events in one command instead of polling and exits after the first event by default. GitHub |
| DEPLOYif you juggle work and personal accounts GitHub Desktop 3.7.0. GitHub Desktop 3.7.0 signs in with more than one account per host, with no fixed limit, and ties each repository to a specific account. GitHub |
| DEPLOYif you deploy to RHEL RHEL runner images. Red Hat Enterprise Linux 9 and RHEL 10 images for GitHub-hosted larger runners are generally available for GitHub Team and GitHub Enterprise Cloud plans. GitHub |
| WATCHif you generate video in Vids Vids Assets section. Google Vids with Gemini Omni now stores generated media in an Assets section for reuse within the same project, rolling out gradually from November 17, 2026. Google Workspace |
| WATCHif you fight bot traffic Priority Projects. Priority Projects are in public beta for Pro and Enterprise teams at $200 per project per month, raising Pro custom rules from 40 to 100 and IP blocks from 40 to 100. Vercel |
| DEPLOYif you track team spend Vercel billing CLI. Vercel CLI now groups billing commands under vercel billing, covering overview, invoice list, inspect, download and upcoming charges, with invoice details in CLI 65.17.0 and later. Vercel |
Left out
- Gemini history import: kept in the Deploy Log; the page carries a four step flow and no figure beyond the feature itself
- Anthropic IPO: a figure in the draft could not be found on the publisher's own page
- CSE Docs collaboration: it did not carry a number that moved
- Workspace Studio Data Regions: it did not carry a number that moved
- Enterprise Live Migrations: it did not carry a number that moved
- Gemini history import: it did not carry a number that moved
- Gemini Tool Retrieval: it did not carry a number that moved
- Gemini computer use on Mac: it did not carry a number that moved
The pattern
The releases this issue converge on one move: the model stops being a thing you call and becomes the runtime your product sits inside. Genie 4 generates continuous 720p worlds at 24fps with synchronized native audio on the Gemini API, and holds world state through Gemini reasoning rather than your own code. GPT-Live 1.1 on Vercel AI Gateway is a full-duplex voice model, and PolyAI Dialog-RSN-1 speaks the OpenAI Realtime protocol with connections up to 30 minutes. Amazon Polly voice cloning builds a cloned voice from 1 to 10 recordings of a consenting speaker in seven languages. The interface layer is being absorbed by the provider.
The second move is that the bill and the control plane are now the product. Amazon Bedrock AgentCore harness runs on Strands harness with lifecycle hooks and proactive context management at about 34% lower token cost than other leading harnesses. The HyperPod Global Inference Gateway cuts a team's 20 model-per-cluster endpoints to one, and SageMaker AI flags endpoints where the busiest GPU used 12 percent or less of capacity. Vercel prices Priority Projects at $200 per project per month and ships vercel billing in CLI 65.17.0 and later. Codex rust-v0.171.0 makes GPT-6.2 Sol the default across 173 commits, and Inkling 2 Mini carries a 1M-token context at 3B active parameters. The teams that win are the ones that read the invoice and the routing table, not the model card.