# AI 24 Updates · 2026-11-19-0600

Published: November 19, 2026 · 06:00 CET

## In Brief

- Google — Gemini 4 Neon launches at $3/$15 per million tokens with a 1M context window
- OpenAI API — 2h and 8h background windows cut token prices by up to 40%
- ChatGPT — ChatGPT for Accounting plugins and Live Dictation
- Claude Code 2.1.328 — auto mode for Opus 4.6 and Sonnet 4.6 on cloud platforms
- Google Labs — Project Genie moves to Genie 4; Stitch Loop opens a waitlist
- Gemini app — text Gemini from Messages or iMessage; runs on lower-memory Android phones
- GitHub Copilot — eight more MCP connectors in public preview
- Perplexity API — pplx-embed-v2-context 9B and 0.6B contextual embedding models
- Perplexity — Teammates and Multiagent arrive on iOS and Android, with voice
- Qwen4-Max — free on Nous Portal for 5 days
- Pika — Enact research preview lets creators direct AI video scenes in real time
- Suno — Suno MCP now available to everyone
- Recraft Studio — attach a typeface to set lettering in generated images
- DiffuSpace — Clover-35B-A3B-Preview technical report and SFT code
- Prime Intellect — Model Factory to train, deploy and improve your own models
- Anthropic — IPO priced at $83 per share, trading on Nasdaq as ANT

---

## 1. Google — Gemini 4 Neon launches at $3/$15 per million tokens with a 1M context window

Company: Google

Google DeepMind released Gemini 4 Neon, a model positioned between Gemini 4 Carbon and Hydrogen for fast, frontier-level software engineering, knowledge work and science tasks.

**Facts:**
- Gemini 4 Neon costs $3 per million input tokens ($0.15 per million cached input tokens) and $15 per million output tokens, with a 1 million-token context window.
- It is rolling out in the Gemini API in Google AI Studio, Gemini for business, Google Antigravity for paid Pro and Ultra subscribers, and the Gemini app for Google AI Pro and Ultra subscribers, where it appears as "Deeper analysis" in Chat and Tasks.
- Google reports a score of 61 on the Artificial Analysis Intelligence Index at the xhigh thinking level, #2 on the Vals Index (70.9%) and 1336 on Arena's Agent Arena Index; the model has 4 adjustable thinking levels.
- GitHub made Gemini 4 Neon generally available in GitHub Copilot for Pro+, Max, Business and Enterprise users, billed at provider list pricing.
- Cursor added Gemini 4 Neon and reports 63.7% on CursorBench 4.1 at default effort, up from 57.3% for Gemini 4 Hydrogen.

**Sources:**
- [Google DeepMind blog](https://deepmind.google/blog/introducing-gemini-4-neon/) (official)
- [GitHub Changelog](https://github.blog/changelog/2026-11-18-gemini-4-neon-in-github-copilot) (official)
- [Cursor on X](https://x.com/cursor_ai/status/2123198346802450926) (official)
- [Gemini app on X](https://x.com/GeminiApp/status/2123169070988337603) (official)

## 2. OpenAI API — 2h and 8h background windows cut token prices by up to 40%

Company: OpenAI

OpenAI opened a public beta of discounted completion windows in Responses API background mode for GPT-6.2 Astra and GPT-6.2 Sol.

**Facts:**
- Setting "completion_window" to "2h" gives 30% off Standard token rates, and "8h" gives 40% off; the default "soonest" has no discount.
- The windows work in Responses API background mode for GPT-6.2 Astra and GPT-6.2 Sol and are in public beta.
- The API targets completion of the first response within the selected window after it accepts the request; the discount also applies to eligible follow-up requests.
- Reusing the window, continuation ID, model and project on follow-ups qualifies them for higher queue priority.
- Separately, OpenAI's deprecations page lists GPT-5.2, GPT-5.4 and GPT-5.4 mini with EU data residency and service_tier fast or priority as deprecated as of Nov 18, 2026.

**Sources:**
- [OpenAI background mode guide](https://developers.openai.com/api/docs/guides/background) (official)
- [OpenAI deprecations](https://platform.openai.com/docs/deprecations) (official)
- [OpenAI Developers on X](https://x.com/OpenAIDevs/status/2123188195202617408) (official)

## 3. ChatGPT — ChatGPT for Accounting plugins and Live Dictation

Company: OpenAI

OpenAI made ChatGPT for Accounting available with a new Accounting plugin and partner plugins, and made ChatGPT dictation faster with Live Dictation.

**Facts:**
- The Accounting plugin adds guidance for invoice processing, reconciliation and reporting, targeting month-end reports, bank reconciliation, data entry and audit prep.
- Partner plugins connect ChatGPT to QuickBooks, Xero, Oracle NetSuite, BILL, FreshBooks, Rillet and Campfire.
- ChatGPT can prepare draft entries from invoices and receipts with suggested account coding and flag unclear details for review.
- Computer use lets ChatGPT help in accounting apps that are not available through a plugin.
- With Live Dictation, speech turns into text while the user is still talking, cutting the wait for a message to finish transcribing.

**Sources:**
- [OpenAI — ChatGPT for Accounting](https://openai.com/chatgpt/accounting/) (official)
- [ChatGPT on X (Accounting)](https://x.com/ChatGPT/status/2123159134963196357) (official)
- [ChatGPT on X (Live Dictation)](https://x.com/ChatGPT/status/2123114715257946334) (official)

## 4. Claude Code 2.1.328 — auto mode for Opus 4.6 and Sonnet 4.6 on cloud platforms

Company: Anthropic

Anthropic's Claude Code v2.1.328 adds auto mode for older Claude models on third-party clouds plus new hook and runner options.

**Facts:**
- Auto mode is added for Claude Opus 4.6 and Sonnet 4.6 on Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry and the Claude apps gateway; interactive sessions with no permission mode start in it.
- mcp_tool hooks gain "onFailure": "block", so a failing hook call blocks the action, as with command and HTTP hooks.
- A new CLAUDE_CODE_DISABLE_DEFERRED_TOOL_RESUME setting lets a host stop claude -p --resume and --continue from running a tool call that a hook deferred.
- The self-hosted runner gains --repository-folder to name the folder sessions use for a repository at --capacity 1.

**Sources:**
- [Claude Code release v2.1.328](https://github.com/anthropics/claude-code/releases/tag/v2.1.328) (official)

## 5. Google Labs — Project Genie moves to Genie 4; Stitch Loop opens a waitlist

Company: Google

Google moved Project Genie onto Genie 4 for eligible Google AI Ultra subscribers and introduced Stitch Loop, an always-on product agent experiment.

**Facts:**
- Genie 4 in Project Genie brings native audio, real-time editing and a new agent for creating interactive virtual worlds.
- Users can prompt mid-session to edit a world live and create worlds conversationally with the agent.
- A community gallery lets users discover, remix and share other users' worlds.
- An updated Genie 3 stays in the model picker with sessions of up to 3 minutes.
- Stitch Loop, a new Google Labs experiment, is an always-on product agent meant to help lean teams build better products; access starts with a waitlist at stitch.google.com/loop/waitlist.

**Sources:**
- [Google DeepMind on X](https://x.com/GoogleDeepMind/status/2123117733709586849) (official)
- [Google Labs on X](https://x.com/GoogleLabs/status/2123118145552749015) (official)
- [Google Labs on X (Stitch Loop)](https://x.com/GoogleLabs/status/2123102389775573314) (official)

## 6. Gemini app — text Gemini from Messages or iMessage; runs on lower-memory Android phones

Company: Google

Google added texting Gemini and its Tasks from Google Messages or iMessage, and re-architected the Gemini app for lower-memory Android devices.

**Facts:**
- After a one-time phone verification at gemini.google.com/channels, users can text Gemini requests and get scheduled check-ins in a private 1:1 thread.
- Replying to the text steers a running task mid-run without starting over.
- Gemini pauses for a 1-tap confirmation before major actions such as sending or booking.
- It is rolling out in English in the U.S. for users 18+ with a personal Google account, a verified U.S. phone number and an eligible carrier.
- The Gemini app now runs on lower-memory Android devices, including eligible Android (Go edition) phones, after Google re-architected it to use less memory.

**Sources:**
- [Gemini release notes](https://gemini.google/release-notes/) (official)
- [Gemini app on X](https://x.com/GeminiApp/status/2123100943604707606) (official)

## 7. GitHub Copilot — eight more MCP connectors in public preview

Company: GitHub

GitHub added eight services that Copilot can connect to through GitHub-verified MCP servers, in public preview.

**Facts:**
- New connectors: Box, Buildkite, Calendly, ClickUp, Mixpanel, Supabase, Webflow and Wix.
- They join more than 25 services already available, such as Slack, Figma and Sentry.
- The connectors are GitHub-verified MCP servers and are in public preview.
- The full list is under Customize in the Copilot app or on the Copilot customization page.

**Sources:**
- [GitHub Changelog](https://github.blog/changelog/2026-11-18-connect-supabase-box-calendly-and-more-to-copilot) (official)

## 8. Perplexity API — pplx-embed-v2-context 9B and 0.6B contextual embedding models

Company: Perplexity

Perplexity added two contextualized embedding models that embed queries, chunked documents and images on POST /v1/embeddings.

**Facts:**
- pplx-embed-v2-context-9b returns 2048 dimensions (or 1024), and pplx-embed-v2-context-0.6b returns 1024 (or 512); the 0.6b model shares the 9b model's 1024-dimension embedding space.
- Documents are sent as an ordered array of chunks, and each chunk's embedding takes the rest of the document into account.
- Page scans or screenshots can be sent as base64 data URIs, with num_chunks giving one embedding per horizontal region; input_type is required.
- Pricing per 1M tokens: 9b costs $0.12 for text and $0.42 for images; 0.6b costs $0.02 for text and $0.40 for images.

**Sources:**
- [Perplexity API changelog](https://docs.perplexity.ai/changelog/changelog) (official)

## 9. Perplexity — Teammates and Multiagent arrive on iOS and Android, with voice

Company: Perplexity

Perplexity brought Teammates and Multiagent to its mobile apps for Max subscribers.

**Facts:**
- Teammates and Multiagent are now in the Perplexity app on iOS and Android.
- Teammates can be controlled by voice.
- Perplexity says about a third of the most demanding Computer work already starts on iPhone.
- Available today for Max subscribers.

**Sources:**
- [Perplexity on X](https://x.com/perplexity_ai/status/2123126865829515363) (official)

## 10. Qwen4-Max — free on Nous Portal for 5 days

Company: Nous Research

Nous Research made Alibaba's flagship Qwen4-Max free in Nous Portal for five days, with Inferact and Alibaba Qwen.

**Facts:**
- Qwen4-Max is free in Nous Portal for 5 days.
- The promotion runs in partnership with Inferact and Alibaba Qwen.
- Nous describes the model as built for long-horizon work, with native vision and 1M context.
- Nous Portal is the account behind the Hermes Agent and lists hundreds of models with free options and Portal-only discounts.

**Sources:**
- [Nous Research on X](https://x.com/NousResearch/status/2123135537087181191) (official)

## 11. Pika — Enact research preview lets creators direct AI video scenes in real time

Company: Pika

Pika shared Enact, a research preview of interactive video scenes built on Pika H3 Flex.

**Facts:**
- Enact is a research preview of scenes that respond to the creator, designed to mimic directing on set.
- It is built on Pika H3 Flex.
- Creators can interact with "actors", manually re-stage a scene and manipulate the camera in real time.
- Access is via a demo request.

**Sources:**
- [Pika on X](https://x.com/pika_labs/status/2123127584342921652) (official)

## 12. Suno — Suno MCP now available to everyone

Company: Suno

Suno opened its MCP server to all users for making songs and visuals inside AI assistants.

**Facts:**
- Suno MCP is now available for everyone.
- Users can make songs from inside ChatGPT, Claude and other AI assistants.
- The same connection also creates visuals.
- Suno's help center adds a Suno MCP article on setup and what users can create with it.

**Sources:**
- [Suno help center](https://help.suno.com/) (official)
- [Suno on X](https://x.com/suno/status/2123119383996457469) (official)

## 13. Recraft Studio — attach a typeface to set lettering in generated images

Company: Recraft

Recraft Studio now sets lettering in a typeface attached to the prompt.

**Facts:**
- Users can attach a typeface from Recraft's Font library or upload their own brand font.
- Recraft Studio sets the lettering in the attached font.
- One font can be attached, or several combined.
- The Font library can be filtered by vibe, appearance or type, from display fonts to serif, monospace and pixel.

**Sources:**
- [Recraft on X](https://x.com/recraftai/status/2123110236718346602) (official)
- [Recraft on X (font library)](https://x.com/recraftai/status/2123129040059933175) (official)

## 14. DiffuSpace — Clover-35B-A3B-Preview technical report and SFT code

Company: DiffuSpace

DiffuSpace Labs published the technical report for its agentic diffusion model Clover-35B-A3B-Preview and open-sourced the SFT code.

**Facts:**
- The report covers the architecture, training recipe and inference experiments behind Clover-35B-A3B-Preview.
- Listed techniques include BD-GDN (recurrent memory with two-way interaction within each block), block-aware expert routing and self-speculative decoding.
- DiffuSpace reports GPU–CPU deployment on a 16GB consumer GPU.
- It reports 73.6 on SWE-bench Verified and 51.7 on Terminal-Bench 2.1.

**Sources:**
- [GitHub SFT code](https://github.com/DiffuSpace/Clover-35B-A3B-Preview/tree/main/sft) (official)
- [DiffuSpace on X](https://x.com/DiffuSpace_Labs/status/2123270249344565377) (official)
- [arXiv](https://arxiv.org/abs/2611.24636)

## 15. Prime Intellect — Model Factory to train, deploy and improve your own models

Company: Prime Intellect

Prime Intellect introduced Model Factory, a single stack for training, deploying and continuously improving custom models.

**Facts:**
- Model Factory puts training, deployment and continuous improvement of a team's own models on one stack.
- Workloads can run on-demand.
- Dedicated clusters are available.
- Customers can bring their own compute.

**Sources:**
- [Prime Intellect on X](https://x.com/PrimeIntellect/status/2123207653530972470) (official)

## 16. Anthropic — IPO priced at $83 per share, trading on Nasdaq as ANT

Company: Anthropic

Anthropic, PBC priced its initial public offering of Class A common stock at $83.00 per share.

**Facts:**
- The offering consists of 980,000,000 Class A shares: 956,351,354 sold by Anthropic and 23,648,646 by selling stockholders.
- Shares are expected to begin trading on the Nasdaq Global Select Market on November 19, 2026 under the symbol "ANT".
- The offering is expected to close on November 20, 2026; underwriters have a 30-day option to buy up to 147,000,000 more shares.
- The SEC declared the registration statement effective on November 18, 2026.

**Sources:**
- [Anthropic news](https://www.anthropic.com/news/anthropic-prices-initial-public-offering) (official)
