Home / New releases / Week of Nov 16–22, 2026

⚡ New AI tool releases: Nov 16–22, 2026

25 verified AI tool, model and product launches from the week of Nov 16–22, 2026, newest first. Each entry links to the official announcement or credible press; dates come from the vendor's post. Last updated 2026-11-20.

Friday, November 20

2026-11-20

Hy4 open weights (Tencent)

Tencent published the open weights for Hy4, the natively multimodal flagship mixture-of-experts model it launched by API on Nov 4: 770B total parameters with 49B active per token (plus a 1.23B vision encoder and a 3.3B DFly2 speculative-decoding drafter), 78 layers, a 1M-token context window, text, image and video understanding, and computer use. The release includes Hy4 (instruct), Hy4-FP8 and the pre-trained Hy4 base model on Hugging Face, ModelScope and GitCode, with GGUF builds from AngelSlim, day-0 vLLM support and a System Card with safety evaluations. It uses the Tencent Hy License 1.0, which is derived from Apache 2.0 but is not Apache 2.0: model-as-a-service businesses with more than US$50 million in annual revenue or products with more than 100 million monthly active users need a separate written license, while purely internal use is exempt.

Source ↗

Thursday, November 19

2026-11-19

Genie 4 in the Gemini API and Google AI Studio (Google)

A day after its Project Genie launch, Google made Genie 4 (genie-4) generally available to developers on the paid tier of the Gemini API (Interactions API, v1beta) and in Google AI Studio. It streams steerable 720p, 24 fps 16:9 worlds with synchronized 48 kHz stereo audio from a text prompt, a 16:9 reference image or both, controlled live with WASD/arrow keys and mid-session text prompts. Pricing is $1.50 per 1M input tokens, $9 per 1M text output tokens and $17.50 per 1M video output tokens (about $0.10 per second of streamed world, audio included); there is no free tier, Batch or Flex option. Every output carries a SynthID watermark.

Source ↗
2026-11-19

Ideogram 4.6

Ideogram released Ideogram 4.6, an image generation and editing model built for precise, multi-step edits, with native output up to 4K (no upscaling). Ideogram says it leaves much less pixel drift outside edited areas, holds color more accurately and renders more realistic texture, and it ranks first on Rapidata's Multi-step Image Edit leaderboard (score 1,892 in the Nov 17 snapshot Ideogram published). It is available in the Ideogram app and the API (Generate and Precise Edit endpoints, from $0.008 per image at Very low quality up to $0.60 for a High-quality 4K edit) and on partner platforms including Luma. Ideogram also said it will release open weights for 4.6 instead of 4.5, without giving a date.

Ideogram on SIloadout → · Source ↗
2026-11-19

GPT-Live-1.1 in the API (OpenAI)

OpenAI made GPT-Live-1.1, the full-duplex voice model that has powered ChatGPT voice for Plus and Pro since Oct 15, available in the API (Live API, v1/live/sessions). Compared with GPT-Live-1 it transcribes speech more accurately, keeps accents and voices more consistent across languages and completes delegated tasks more reliably. A new standby option for GPT-Live-1 and GPT-Live-1.1 WebRTC sessions pauses inference after 5 to 60 seconds of silence and bills $0.02 per minute instead of the $0.05 active rate; backend model and tool use is billed separately.

Source ↗
2026-11-19

Aion Instruct open weights and Edge Prompt API origin trial (Microsoft)

Microsoft made Aion Instruct (Aion-1.0-Instruct-2.3B), a 2.3B-parameter on-device small language model from its Windows Applied Sciences group, generally available as open weights on Hugging Face under the MIT license and as Microsoft Edge's built-in language model, replacing Phi-4-mini. It handles summarization, rewriting, chat and structured output in eight languages with a 32K-token context. Edge also opened an origin trial of the Prompt API so live sites can use the model on-device, plus a developer preview of an experimental Decisions API. Microsoft says Aion Instruct is also coming to Windows.

Source ↗
2026-11-19

Theorizer in Asta, open source (Ai2)

Ai2 launched Theorizer in Asta, its platform of AI agents for scientific research. Theorizer searches the literature, extracts findings from papers and combines them into testable theories, linking each claim to the supporting studies. Ai2 also open-sourced the code (asta-theorizer, Apache 2.0) so researchers can run it locally with their own model API keys; in Asta, Ai2 covers token costs up to a limit. Scientists have tested it for nearly a year, and Ai2 cautions that it can miss relevant studies and misjudge originality.

Source ↗
2026-11-19

Olala mini 7A1B open weights (OVHcloud)

OVHai LLM (OVHcloud group) released Olala-v1.0-mini-7A1B, the first model in its Olala family, with open weights under Apache 2.0 on Hugging Face. It is a small reasoning model for chat, instruction following and tool use (6.8B total parameters, about 1B active per token, hybrid Mamba-3 / attention mixture-of-experts, 32K context with a 128K version coming soon), trained on three European public supercomputers on 12.8T tokens, and it can run on a laptop or CPU. Base, SFT, summarise and translate variants were released alongside it, and it is hosted on OVHcloud AI Endpoints and Hugging Face Inference Providers.

Source ↗

Wednesday, November 18

2026-11-18

Hy4 flash open weights (Tencent)

Tencent released open weights for Hy4 flash, the low-cost everyday model it launched by API on Nov 16: Hy4 flash, Hy4 flash-FP8 and Hy4 flash-Base on Hugging Face and GitHub, with GGUF builds from AngelSlim and day-0 vLLM support. The model card lists 220B total parameters (including 64B Engram parameters) with 12B active per token, native text and image input and a 1M-token context window. It is released under the Tencent Hy License 1.0, which is derived from Apache 2.0 but adds a separate-license requirement for large model-as-a-service businesses.

Source ↗
2026-11-18

Gemini 4 Neon (Google)

Google introduced Gemini 4 Neon, a model that sits between Gemini 4 Hydrogen and Gemini 4 Carbon, aimed at complex software engineering, knowledge work and science tasks with fast end-to-end completion. It has a 1M-token context window and four thinking levels (low, medium, high by default, xhigh) and costs $3 per 1M input tokens ($0.15 cached) and $15 per 1M output tokens. It is rolling out in the paid Gemini API in Google AI Studio, Gemini for business, Google Antigravity (paid AI Pro and Ultra) and the Gemini app for AI Pro and Ultra subscribers, and is also generally available in GitHub Copilot and Cursor.

Source ↗
2026-11-18

Genie 4 in Project Genie (Google DeepMind)

Google DeepMind made Genie 4, its most realistic interactive world model, available to eligible Google AI Ultra subscribers in Project Genie (an experimental browser prototype; AI Pro members can join a waitlist). Users can type prompts mid-session to change the world in real time, and Genie 4 adds real-time native audio for the first time. The update also brings a Project Genie agent that builds worlds from a conversation, a community gallery for remixing worlds, and Genie 3 upgrades (up to 3 minutes per generation, world chaining and 360-degree panorama seeding). Output is real-time 720p at up to 24 fps, with SynthID watermarking.

Source ↗
2026-11-18

Suno MCP for all plans (Suno)

Suno made Suno MCP available to everyone, including Free-plan users, after an early-access beta. Adding https://api.suno.com/mcp as a custom connector lets ChatGPT, Claude, Claude Code, Gemini, Cursor and other MCP clients create songs, covers, extends and mashups, split stems, convert to MIDI, make sounds and manage the Suno library from the chat, with no API key. Creations use the account's Suno credits and save to the library; Visuals (art and looping video) need a paid plan.

Suno on SIloadout → · Source ↗
2026-11-18

AssemblyAI-STT-3.7

AssemblyAI released AssemblyAI-STT-3.7, a new generation of speech-to-text models in three versions: AssemblyAI-STT-3.7-Realtime for live meetings, calls and captions ($0.30/hr), AssemblyAI-STT-3.7-Realtime-Agent for voice agents ($0.45/hr; focuses on the person talking to the agent, with final transcripts in about 86 ms) and AssemblyAI-STT-3.7 for recorded audio ($0.21/hr). All three cover 32 languages with automatic detection (99 with the Universal-2 fallback on pre-recorded audio), and realtime speaker labels are final as soon as a turn ends. They replace Universal-Streaming, Universal-3.6 Pro Realtime and Universal-3.5 Pro; Universal-3.5 Pro Realtime retires on February 1, 2027.

Source ↗

Tuesday, November 17

2026-11-17

Replit Missions (Replit)

Replit launched Missions: you define an outcome, success criteria, a spending cap and which actions need your approval, and Replit coordinates a swarm of agents to plan, build, test and iterate toward it, with visible progress and an independent AI judge checking results against the agreed criteria. Missions are started from the New menu. Replit's own examples: agents built an internal application-monitoring platform in two days within a $3,000 budget, and 32-agent swarms reached up to 19x speedups on NumPy benchmark tasks (company figures).

Replit on SIloadout → · Source ↗
2026-11-17

GPT-6.2 Astra and GPT-6.2 Sol (OpenAI)

OpenAI released GPT-6.2 Astra, its new frontier model for the hardest professional, coding and science work, and GPT-6.2 Sol, an efficient model for everyday work; both use fewer output tokens per task. They are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex (Ultrafast for Pro 500, and for Pro 200 starting with Sol) and in the API as gpt-6.2-astra and gpt-6.2-sol. API prices per 1M tokens: Astra $8 input, $0.40 cached input and $40 output (20% below GPT-6 Astra); Sol keeps GPT-6.1 Sol's $2 / $0.10 / $10.

Source ↗
2026-11-17

MAI Code 2 and MAI Code 2 Flash in GitHub Copilot (Microsoft AI)

Microsoft AI launched MAI Code 2 and MAI Code 2 Flash at Ignite, two coding models aimed at everyday enterprise work (multi-file changes, debugging, UI and dashboards; Flash for quick edits, explanations and routine fixes) rather than very long-horizon agentic tasks. Both are available in GitHub Copilot, and Code 2 Flash is in limited-access preview in Microsoft Foundry, with OpenRouter, Vercel, OpenCode and Kilo listed as coming soon. Microsoft's own benchmark claim: Code 2 reaches 98% of GPT-6.1 Sol's quality at about 40% of the price (launch costs shown at a 50% promotional price).

GitHub Copilot on SIloadout → · Source ↗
2026-11-17

MAI Image 2.7 and MAI Image 2.7 Flash in Microsoft Foundry (Microsoft AI)

Microsoft AI made MAI Image 2.7, its image editing and text-to-image model (previewed in October), and the new MAI Image 2.7 Flash for high-volume generation at about $10 per 1,000 images available in Microsoft Foundry, as part of seven MAI models announced at Ignite. Microsoft says MAI Image 2.7 is #1 on Artificial Analysis's image-editing leaderboard and #4 for text-to-image. The same announcement covers MAI Cyber 2 Flash (restricted access), MAI Realtime 1 (register interest) and Decision 1.1.

Source ↗
2026-11-17

Microsoft-Decision-1.1 GA (Microsoft)

Microsoft made Microsoft-Decision-1.1 generally available, a fast, low-cost decision model for structured choices in software such as routing, classification, prioritization, verification and workflow control. It picks among developer-permitted actions, returns calibrated confidence scores so apps can defer when unsure, and adds multimodal support over Decision 1. It is available in Microsoft Foundry and through OpenRouter and Vercel AI Gateway; Microsoft says it is 50x faster and 79x cheaper than GPT-6.1 Sol.

Source ↗
2026-11-17

Kling Image 4.0 (Kling AI)

Kling AI released Kling Image 4.0, its next-generation image model built on a unified understanding-and-generation autoregressive architecture, live in Kling Creative Studio for all users and via API. It generates native 8K images, renders up to 2,000 characters of text per image, makes sets of up to 10 consistent images from one prompt and long-form images for e-commerce product pages and web design. API pricing starts at $0.063 per 2K image ($0.094 at 4K, $0.157 at 8K), and reference images are now billed at $0.003 each.

Kling AI on SIloadout → · Source ↗
2026-11-17

Devin Bots (Cognition)

Cognition launched Devin Bots, persistent engineering agents with their own memory and tools that own an ongoing goal (for example maintaining a repo, keeping an Android app at parity with iOS or managing releases), start and follow up on parallel Devin sessions, and bring back only the decisions that need a human. Bots can run on schedules or tool triggers and report in Slack; Cognition says its own Android Bot built Devin's Android app. Available now at devin.ai/bots.

Source ↗
2026-11-17

Lumo Desktop with Connectors, Schedules and Skills (Proton)

Proton released Lumo Desktop, a desktop app for its private AI assistant on Windows, macOS and Linux, and added Connectors, Schedules and Skills. Connectors (in the desktop app) let Lumo use local tools (filesystem access, shell execution and browser content fetching), Proton Mail through Proton Mail Bridge, Google Drive, Jira, Confluence and your own local or remote MCP servers; for each action you choose whether it runs automatically, needs approval or is blocked. Schedules run tasks at a set time or on repeat, and Skills save reusable instructions (Custom Lumos are being migrated to Skills). Proton says chats are zero-access encrypted and never used for training, and the source code is public.

Source ↗
2026-11-17

HeyGen Avatar 6

HeyGen's new avatar model, publicly launched Nov 17 after early access for enterprise workspaces. It renders every frame in native 2K (export in 1080p or 2K), reads the script to match gestures such as pointing and counting, and can put two avatars in one conversation. HeyGen reports that side by side with real footage it is 84.4% indistinguishable, up from 44.4% for Avatar V (HeyGen's own evaluation). An avatar of yourself needs a photo plus 30 seconds of video. It works in the app, the API and HeyGen MCP; the first video is free, then it uses credits, with plans from $29/month for 600 credits (Avatar 6 Turbo on Creator, Avatar 6 Pro on Pro).

HeyGen on SIloadout → · Source ↗

Monday, November 16

2026-11-16

Qwen4-Max (Alibaba)

The production release of Alibaba's flagship Qwen model, following the Nov 11 Qwen4-Max-Preview. It is a mixture-of-experts model with 2.4 trillion total parameters (71B active per token), native vision and a 1M-token context window. Alibaba reports it ranks 6th on Code Arena and beats Qwen3.8-Max on its own Work360 benchmark while using 35% fewer tokens at high effort and cutting the average cost per task by about 60%. It is available now through the QwenCloud API and Token Plan and in QwenWork and QwenStudio. Model Studio (Singapore) lists $1.50 per 1M input and $4.50 per 1M output tokens, with a 1M-token free quota for 90 days. Smaller open-weight Qwen4 models are coming soon.

Source ↗
2026-11-16

Hy4 flash (Tencent)

Tencent's lowest-cost Hy model so far: 224B total parameters (including 64B Engram parameters) with 12B active per token, natively multimodal with a 1M-token context window. API pricing is $0.07 per 1M input, $0.281 per 1M output and $0.003 per 1M cached input tokens. In an internal blind test with 85 WorkBuddy users, Tencent says it scored close to Claude Haiku 5.5 on satisfaction, and on its WorkIndex benchmark it costs about 1/20 as much per task as Hy4. It is available now in Tencent Cloud TokenHub (50% off through Nov 30), WorkBuddy (free through Nov 30), Yuanbao, WeCom, QQ and more. OpenRouter availability and open weights are planned.

Source ↗
2026-11-16

Monitor API (Perplexity)

A new Perplexity API that watches the web for new developments on a topic, or tracks changes to up to 10 specific pages, on a schedule (every 15 minutes, hourly, daily, weekly or custom cron) and sends relevant updates to signed webhooks. Perplexity Decider checks each web result for relevance, source legitimacy, new information and freshness, with adjustable thresholds; page monitors return a text diff of what changed. Web monitors cost $1 per 1,000 successful runs and page monitors $1 per 1,000 URLs fetched. Available now through the API and API Console with the same key as the Search, Context and Agent APIs.

Perplexity on SIloadout → · Source ↗
2026-11-16

d1.2 and d1.2-ultra decision models (Liquid AI)

New versions of Liquid AI's decision models, which answer structured questions with probabilities. d1.2 and d1.2-ultra add audio input (alongside text and images) and five new answer types: reading a value, finding every passage of a kind, locating objects with bounding boxes, estimating a number with a range, and picking every option that applies. Liquid reports d1.2-ultra scores 75, first among decision models on the Artificial Analysis Decision Index v2.1. Both are available now on the Liquid API, billed on input tokens only: $0.02 per 1M for d1.2 and $0.042 per 1M for d1.2-ultra. OpenRouter and Vercel support is coming soon.

Source ↗

Frequently asked questions

What AI tools launched the week of Nov 16–22, 2026?

Hy4 open weights (Tencent), Genie 4 in the Gemini API and Google AI Studio (Google), Ideogram 4.6, GPT-Live-1.1 in the API (OpenAI), Aion Instruct open weights and Edge Prompt API origin trial (Microsoft), Theorizer in Asta, open source (Ai2), Olala mini 7A1B open weights (OVHcloud), Hy4 flash open weights (Tencent), Gemini 4 Neon (Google), Genie 4 in Project Genie (Google DeepMind), Suno MCP for all plans (Suno), AssemblyAI-STT-3.7, Replit Missions (Replit), GPT-6.2 Astra and GPT-6.2 Sol (OpenAI), MAI Code 2 and MAI Code 2 Flash in GitHub Copilot (Microsoft AI), MAI Image 2.7 and MAI Image 2.7 Flash in Microsoft Foundry (Microsoft AI), Microsoft-Decision-1.1 GA (Microsoft), Kling Image 4.0 (Kling AI), Devin Bots (Cognition), Lumo Desktop with Connectors, Schedules and Skills (Proton), HeyGen Avatar 6, Qwen4-Max (Alibaba), Hy4 flash (Tencent), Monitor API (Perplexity), d1.2 and d1.2-ultra decision models (Liquid AI).

How does SIloadout verify releases?

Each launch is checked against the vendor's own post, its official X account or credible press before it is added, and every source link is confirmed to load. Rumored, unreleased or ambiguously dated items are left out.

Older: Nov 9–15, 2026 → · All releases · Atom feed · JSON Feed

Browse by week: Nov 16–22, 2026 (25) · Nov 9–15, 2026 (38) · Nov 2–8, 2026 (20) · Oct 26–Nov 1, 2026 (30) · Oct 19–25, 2026 (45) · Oct 12–18, 2026 (33)

Browse by month: November 2026 (83) · October 2026 (108)

Open models only: New open-weight and open-source AI releases (33)

By company: Google (10) · OpenAI (7) · Microsoft (6) · Perplexity (5) · Tencent (5) · Anthropic (4) · ElevenLabs (4) · NVIDIA (4) · Baidu (3) · GitHub (3) · HeyGen (3) · Meta (3) · MiniMax (3) · Mistral AI (3) · Moonshot AI (Kimi) (3) · All companies