Gemini 3.8 Live Avatar
Gemini 3.8 Live with Live Avatar adds near real-time visual presence to live dialogue, with lip-syncing, expressions, and turn-taking across 97 languages. It is available in Gemini Enterprise starting today.
What shipped in AI, week of September 21, 2026
134 deployments dated September 21 to 27, 2026 are in The Deploy Log: 3 leads, 9 in Models and APIs, 121 in Tools and products and 1 in Robotics, hardware and chips, from 42 publishers. They include Gemini 3.8 Live Avatar from Google, Gemini 3.8 TTS from Google, Gemini Omni in Vids from Google, MiMo-V2.6-Flash-MOPD and MiMo-V2.6-Pro-MOPD from Xiaomi MiMo, Health Guardian features from Google and 129 others. Every item sourced to the publisher's own page.
Leads
Gemini 3.8 TTS
Gemini 3.8 Flash TTS and Flash-Lite TTS generate speech in over 100 languages, with voice creation from natural language prompts, 2,000+ production-ready voices, and voice replication from a 30-second sample. Both are rolling out today in the Gemini API and Google AI Studio.
Gemini Omni in Vids
Google Vids now lets anyone with a Google or Google Workspace account generate 1080p HD videos at no cost using Gemini Omni 1.1 Flash, with scene extension, exact clip durations, and upscaling of existing AI clips.
Models and APIs
MiMo-V2.6-Flash-MOPD and MiMo-V2.6-Pro-MOPD
MiMo-V2.6-Flash-MOPD and MiMo-V2.6-Pro-MOPD released on Hugging Face with 1M token context and 309B total parameters.
Health Guardian features
Health Guardian features are rolling out in the Google Health app, with blood pressure trends, insulin resistance trends, and sleep breathing quality metrics included free with Pixel Watch 3, 4, and 5 devices.
Intern-Decision-4B
InternLM releases Intern-Decision-4B, 2B, and 0.8B multimodal structured decision models on Hugging Face.
Perceptron Mk1.5
Perceptron Mk1.5 is available on OpenRouter at $0.15 input and $1.50 output per 1M tokens.
TypeSafe Jev Router
Jev Router is available on OpenRouter with a 1,000,000 token context window and zero pricing.
ChatGPT Ads
ChatGPT Ads now available in more than 60 countries after adding Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam, and Taiwan.
Gemini Connected Apps
A new wave of Connected Apps is rolling out to Gemini.
Claude Opus 5.5 on AWS
Claude Opus 5.5 is now available on AWS.
Grok 4.6 in Amazon Bedrock
xAI's Grok 4.6 is now available in Amazon Bedrock.
Tools and products
Compute Engine X5 series
Compute Engine X5 series is generally available with 48TB memory configurations, up from the X4's 32TB limit.
Copilot CLI v1.0.89-5
Copilot CLI v1.0.89-5 pre-release adds Claude Code rule files in .claude/rules as custom instructions.
InvokeAI 6.14.2
Version 6.14.2 adds custom font import for Canvas text and style transfer to krea-2 via reference image.
KoboldCpp 1.122.1
KoboldCpp 1.122.1 adds an integrated KoboldCpp Agent with 9 built-in tools and a 2k token system prompt.
Lovable connectors and GPT-6 models
Lovable adds Discord and Google Business Profile connectors, plus GPT-6 Sol and GPT-6 Luna for app AI features.
Unsloth CUDA 13 Wheels
Prebuilt Linux x86_64 CUDA 13 wheels for flash-attn 2.8.4, causal-conv1d 1.7.0, and mamba-ssm 2.3.2.post1.
Codex rust v0.157.0
Codex rust-v0.157.0 adds GPT-6 Sol and Luna to the model catalog, including Amazon Bedrock support and migration prompts for older models.
Zed v1.21.0
Zed v1.21.0 adds BYOK support for Claude Opus 5.5 with an Anthropic API key and GPT-6 Astra, Sol, and Luna with an OpenAI API key.
Ollama v0.34.4
Ollama v0.34.4 makes structured outputs on thinking models apply in a single pass, and Qwen 3.8 prompt processing is faster on Apple Silicon.
Copilot weekly releases
GitHub Copilot adds Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7 across Pro, Pro+, Max, Business, and Enterprise plans, with local sandboxing in the Copilot app now in public preview.
Gemini TTS on AI Gateway
Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Flash TTS from Google are now available on AI Gateway, generating speech in more than 100 languages with long-form narration and two-speaker dialogue.
Omni 1.1 in Vids
Google Vids now supports generating and upscaling AI videos to 1080p with Gemini Omni 1.1 Flash, with full rollout starting September 23, 2026.
Meet occupancy counting
Occupancy counting is now available for Android-based Logitech Gen 2 room hardware, starting with Logitech Rally Board 65, with an extended rollout starting September 28, 2026.
Compute Engine X5 48TB instances
New X5 series generally available with 48TB memory configurations, up from the 32TB X4 limit.
PostHog AI Observability Backfill
AI Observability evaluation can now run on data already collected, instead of only on new data as they arrive.
16k context LLM models
RunPod now supports Panchovix's 16k-token context models.
AI agent skills for AWS MCP Server
AWS End User Messaging and Amazon SES now offer AI agent skills for the AWS MCP Server.
Better Forge template
RunPod released Better Forge, a template for Stable Diffusion pods with a total payload of approximately 12GB.
Claude Agent SDK TypeScript v0.3.283
v0.3.283 adds plugin_errors to SDKSystemMessage, fixes session rewinding, and includes warnings in stream-json output.
Cline Desktop v0.0.36
Desktop v0.0.36 fixes proxy startup failures and window sizing on scaled displays.
DSPy 3.4.0
DSPy 3.4.0 introduces Jev integration with three experimental decision types and native LM engines.
EC2 C8i and C8i-flex instances
Amazon EC2 C8i and C8i-flex instances are now available in additional regions.
EC2 M8i and M8i-flex instances
Amazon EC2 M8i and M8i-flex instances are now available in additional regions.
EC2 R8i and R8i-flex instances
Amazon EC2 R8i and R8i-flex instances are now available in additional regions.
FastMCP v4.0.10
Task-enabled tools now work behind search transforms and CodeMode.
Flutter Replay Screenshots
Session Replay for Flutter Android can use configurable screenshot resolution, JPEG quality, and native-screen pixel formats.
Genie Code document processing
Genie Code can now build end-to-end document processing pipelines by orchestrating parsing, classification, extraction, and validation.
Genie One MCP server
The Genie One MCP server is now generally available as a Databricks-provided MCP Service in Unity Gateway.
GitHub Actions query result cap
GitHub Actions API and UI now report "2,500+" when workflow run query results exceed 2,500 records.
GitHub Copilot CLI 1.0.89-4
Copilot CLI pre-release 1.0.89-4 adds Auto routing tier suggestions and configurable direct plugin installs.
Google Calendar time zones
Google Calendar on the web now displays up to three time zones, up from two.
Kilo Code JetBrains 7.1.8
JetBrains 7.1.8 adds checkpoints settings, companion skill installs, CycloneDX SBOMs, and OAuth client ID metadata support.
Kilo Code v7.8.1
Publishes CycloneDX SBOMs and supports public HTTPS pages in localhost previews.
LangSmith Custom Apps
LangSmith Custom Apps lets users build, publish, and run custom UIs on LangSmith data directly inside LangSmith.
Langfuse v4.45.4
v4.45.4 refreshes embedded Langfuse skills and skips Azure LLM connection tests.
Langfuse v4.46.0
v4.46.0 adds side-by-side experiment comparison, formatted JSON, skill management, and 100-row tracing tables.
LiteLLM v1.100.3
LiteLLM v1.100.3 released with 7445 commits to main since this release.
Memorystore for Valkey 9.1
Memorystore for Valkey 9.1 is generally available with up to 3x QPS at microsecond latency.
Nemotron 3 Diarization on DigitalOcean
NVIDIA Nemotron 3 Diarization model is now available on DigitalOcean Inference for serverless inference and Agent Development Kit.
OpenHands v1.24.0
v1.24.0 adds toggling all workspace folders, scoped OpenAI subscription caches, and read-only shared automation conversations.
Pixel Canary on AI Gateway
Pixel Canary is available as stealth/pixel-canary on AI Gateway, free for a limited time, with 90.3% Next.js eval baseline.
Pydantic AI v2.50.0
Adds DecisionModel and exposes OpenAI service_tier in provider details.
Pydantic AI v2.51.0
Adds OpenAI GPT-Live support with OpenAILiveModel and exposes context_window_used.
Railway Free VMs
Free VMs with 2 vCPU and 2 GB RAM are available via ssh railway.new with no account required.
Raycast External Model Providers
Raycast v2.5 lets extensions provide their own AI models to AI Chat.
RunPod Serverless GPU and SGLang
RunPod now supports up to four 80GB GPUs per worker and offers an SGLang quick deploy endpoint.
Storage Intelligence advisor
Storage Intelligence advisor is generally available along with expanded storage batch operations capabilities.
Tags for Databricks assets
Tags for dashboards, notebooks, Databricks apps, and Genie Agents are now generally available.
Vercel Container Registry GitHub Action
New vercel/vcr-action/login action authenticates with GitHub OIDC and pushes container images to VCR.
Vercel Sandbox memory observability
Sandbox memory usage data now available in dashboard and CLI, reporting average, P75, and P95 memory.
runpod-python 0.10.0 Local API Server
runpod-python 0.10.0 adds a local API server for testing worker handlers via --rp_serve_api.
xAI Grok 4.7 in OCI Generative AI
xAI Grok 4.7 is now available on demand in OCI Generative AI with a 500,000-token context window.
Amazon GameLift Servers
GameLift Servers is now available in 5 new regions and 8 Local Zones.
Amazon RDS for MySQL Extended Support
Amazon RDS for MySQL Extended Support adds minor versions 5.7.44-rds.20260902 and 8.0.46-rds.20260908.
Amazon RDS for PostgreSQL post-quantum TLS
Amazon RDS for PostgreSQL now supports post-quantum TLS key exchange.
Android Studio BYOA
Android Studio Canary now supports Bring Your Own Agent with Claude Agent, OpenAI Codex, and Google Antigravity.
Azure Instant Access for VM restore points
Azure Instant Access for VM restore points is generally available.
Model Armor Melbourne
Model Armor prompt injection and jailbreak detection filters are available in Melbourne with data residency enforcement.
harness-cli v0.1.3
harness-cli v0.1.3 replaces setup Q&A with Quickstart, Customize, and Export, and adds /tools.
huggingface_hub v1.33.0
huggingface_hub v1.33.0 adds a validating progress bar for large Xet uploads and simpler skills installs.
huggingface_hub v2.0.0
huggingface_hub v2.0.0 replaces its HTTP dependency with httpx2.
langchain-openai 1.6.6
langchain-openai 1.6.6 released with a fix that raises on error events in the stream path.
ComfyUI v0.37.2
ComfyUI v0.37.2 released with no specific numeric change stated.
Connect routing step data in lake
Amazon Connect Customer now provides routing step data in the analytics data lake.
Gemini Notebook in Google Docs
Google Docs can now ground AI prompts in existing Gemini Notebook sources with inline citations.
Gemini Omni 1.1 Flash in Vids
Gemini Omni 1.1 Flash is now in Google Vids with 1080p generation and upscaling, rolling out starting September 23, 2026.
GitHub Actions Node 20 retirement
GitHub Actions runners now use Node 24, and Node 20 is no longer available.
GitHub Copilot Local Sandboxing
GitHub Copilot app adds per-project local sandboxing for filesystem, network, and credential access, off by default.
GitHub Copilot OTel support
GitHub Copilot app now supports OpenTelemetry configuration through enterprise-managed settings.
Google Admin temporary roles
Admins can assign roles for predefined durations such as 30 days or a custom expiration date and time.
Google Meet occupancy counting
Occupancy counting is available for Android-based Logitech Gen 2 room hardware starting with Rally Board 65.
LiteLLM v1.100.2
LiteLLM released version 1.100.2.
LiteLLM v1.101.1
Released LiteLLM v1.101.1.
LiteLLM v1.102.1
LiteLLM v1.102.1 released with 4148 commits since previous release.
LiteLLM v1.99.3
LiteLLM released version 1.99.3.
Vercel Blob stores
Store creation limits of 100/500/1,000 are removed; creation is billed at $5.00 per million on Pro.
claude-code v2.1.281
Claude Code v2.1.281 adds gateway support for desktop policy blocks, Bedrock IAM role assumption, and Bedrock guardrails.
codex rust-v0.156.1
Release 0.156.1 of the codex Rust crate is tagged.
langchain-anthropic 1.7.4
langchain-anthropic 1.7.4 adds Opus 5.5 and GPT-6 profile augmentations.
langchain-openai 1.6.5
langchain-openai 1.6.5 adds Opus 5.5 and GPT-6 profile augmentations and mid-conversation tool changes.
llama.cpp v0.5.0
llama.cpp v0.5.0 adds HRM-Text support, MiMo-V2.6 and HunyuanOCR conversion, and multi-address HTTP binding.
vLLM v0.30.1rc0
vLLM v0.30.1rc0 adds MI355 dense NVFP4 and MoRI kernel mirrors.
Amazon EMR 7.14
Amazon EMR 7.14 is now available.
C++ whole codebase indexing
C++ code intelligence in GitHub Copilot CLI now supports whole codebase indexing, enabled by default.
Claude Opus 5.5
Claude Opus 5.5 is now the default Opus model with 1M context and $4/$20 per Mtok.
Claude Opus 5.5 in GitHub Copilot
Claude Opus 5.5 is available in GitHub Copilot and billed at provider list pricing under usage-based billing.
Claude Opus 5.5 on AI Gateway
Claude Opus 5.5 is available on AI Gateway with a 1M-token context window and 128K max output tokens.
Claude Opus 5.5 on AWS GovCloud
Claude Opus 5.5 is now available on AWS GovCloud (US).
ComfyUI v0.37.1
ComfyUI v0.37.1 tagged on 22 Sep with no numbered change listed.
Confluence for Google Chat
Confluence integration for Google Chat is available now to all Google Workspace customers.
Drives for Vercel Sandbox
Drives for Vercel Sandbox are now in public beta on Hobby, Pro, and Enterprise with a default maximum size of 1 TiB.
GPT-6 Sol and GPT-6 Luna
OpenAI GPT-6 Sol and GPT-6 Luna are now generally available on Amazon Bedrock.
GPT-6 Sol and Luna in Copilot
GPT-6 Sol and GPT-6 Luna are now selectable in GitHub Copilot across Pro, Pro+, Max, Business, and Enterprise plans.
GPT-6 Sol and Luna on AI Gateway
GPT-6 Sol and GPT-6 Luna are now available on Vercel AI Gateway at a lower price than GPT-6 Astra.
Genie One MCP
Genie One MCP is now generally available.
Google Sheets pivot table sorting
Google Sheets pivot tables now support manual reorder and custom sort, available to all Workspace customers.
LiteLLM v1.102.0
LiteLLM v1.102.0 tagged with no numbered change listed.
Mac mini and Mac Studio
Mac mini with M6 starts at $899, Mac Studio with M5 Ultra starts at $5,499, with up to 4.3x faster AI performance.
NVIDIA P100 EOS on Compute Engine
NVIDIA P100 GPUs on Compute Engine reached end of support and are shut down.
Ollama v0.34.3
Ollama v0.34.3 adds Nemotron H vision model support on Apple Silicon with MLX and fixes HuggingFace model pulls.
langchain-anthropic 1.7.3
langchain-anthropic 1.7.3 released with structured output routing fix and mid-conversation system message support.
langchain-deepseek 1.1.1
langchain-deepseek 1.1.1 released with dependency bumps and strict mode routing fix.
langchain-fireworks 1.6.2
langchain-fireworks 1.6.2 released with test fixes and dependency bumps.
langchain-openai 1.6.4
langchain-openai 1.6.4 released with refreshed OpenAI model profile data.
langchain-openrouter 0.2.9
langchain-openrouter 0.2.9 released with dependency bump anyio 4.13.0 to 4.14.2.
vLLM v0.30.0
vLLM v0.30.0 fixes DeepGEMM CUDA 12.9 release builds.
AI Gateway Jev support
AI Gateway now supports Jev through TypeSafe clients, HTTP API, and AI SDK.
Amazon EVS FedRAMP Class C
Amazon EVS is now in scope for FedRAMP Class C.
Apigee X maintenance
Apigee instances below 1-18-0-apigee-4 will be updated within seven to 21 days.
Azure Sphere OS 26.09
Azure Sphere OS version 26.09 is now generally available.
ComfyUI v0.37.0
ComfyUI v0.37.0 is now available.
GitHub Enterprise credential inventory exports
Enterprise owners can export all credentials via CSV or paginated REST API.
Google Sheets manual calculation
Editors can pause automatic recalculation and trigger a sheet-wide update manually.
Grok 4.7 on Vercel AI Gateway
Grok 4.7 is available on AI Gateway with a 500K token context window and 40% off through September 27.
MiMo V2.6 on AI Gateway
MiMo V2.6 Pro, Flash, and Pro UltraSpeed are available with 1M token context and up to 128K output tokens.
Open WebUI v0.11.4
Slim image is now about 175 MB, 89% smaller than the last release.
Python Workers
Python Workers are now generally available on Cloudflare.
Vercel Connect Microsoft Teams
Vercel Connect now includes a managed connector for Microsoft Teams, available from the dashboard or CLI.
Workday for Google Sheets
Workday for Google Sheets is a new add-on available in the Google Workspace Marketplace.
langchain-core 1.6.4
langchain-core 1.6.4 released with deprecation of chat message history.
langchain-openai 1.6.3
langchain-openai 1.6.3 released with support for GPT-6 request constraints.
Robotics, hardware and chips
Isaac ROS 5.0
Isaac ROS 5.0 released with support for ROS Lyrical and Ubuntu 24.04.
The edition every row came from arrives in your inbox, free.
Join free