← All articles
Roundup

What changed in AI tools this week: GPT-6.2, Gemini 4 Neon and free Gemini Tasks

A practical roundup of AI tool updates from November 14 to 18, 2026: new OpenAI and Google models, a free Gemini task agent, new privacy controls, Hugging Face routing, and creative tools from Snap, Google DeepMind and Suno.

The past week was heavy on Google, but the updates that matter most for everyday users are spread across several companies. OpenAI refreshed its two main models, Google added a mid-priced Gemini model and made its task agent free, two privacy controls arrived for Gemini and Search, Hugging Face changed how it routes requests to open-weight models, and creative tools from Snap, Google DeepMind and Suno all picked up new AI features.

This roundup covers announcements made between November 14 and November 18, 2026. Each section says what changed, who can use it, and what it means if you are choosing or already paying for one of these tools. All facts come from the companies' own announcements or from named news reports, linked in each section. We have not tested these features ourselves; where a company makes a performance claim, we say whose claim it is.

OpenAI: GPT-6.2 Astra and Sol

What happened. On November 17, OpenAI released two new models, GPT-6.2 Astra and GPT-6.2 Sol. According to The Decoder, Astra is meant for demanding specialist work in areas such as science, law and finance, while Sol is the quick model for everyday tasks.

Who gets it. The Decoder reports that both models are available to Plus, Pro, Enterprise and Edu subscribers in ChatGPT Work and in Codex, and to developers through the OpenAI API. Pro-500 subscribers can use an Ultrafast mode with both models, and Pro-200 subscribers get Ultrafast for the first time, starting with Sol.

Prices. For the API, The Decoder lists Astra at $8 per million input tokens, $0.40 for cached input and $40 per million output tokens, which OpenAI says is 20 percent cheaper than the previous Astra. Sol keeps the earlier Sol price of $2 for input, $0.10 for cached input and $10 for output, per million tokens.

Claims worth knowing. OpenAI says Astra follows professional conventions better, for example by taking layouts, colours and typography from a template when it builds slides. On the independent Artificial Analysis Intelligence Index, The Decoder reports that Astra ranks second, behind Anthropic's Claude Fable 5.2.

What it means for you. If you pay per token, a cheaper top model is a reason to re-run your own tests, especially for document and slide work. For subscribers on the plans listed above, both models are already available, so it is worth trying Sol for routine work before reaching for Astra.

Google: Gemini 4 Neon fills the middle of the lineup

What happened. On November 18, Google introduced Gemini 4 Neon, a model aimed at fast, complete work on software engineering, office tasks and science. It joins Gemini 4 Hydrogen, Carbon and Argon.

Who gets it. Google says Neon is rolling out in the Gemini API in Google AI Studio, in Gemini for business, in Google Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers.

Prices. Google lists $3 per million input tokens and $15 per million output tokens, with cached input at $0.15 per million. The context window is 1 million tokens.

Claims worth knowing. Google says Neon scores 61 on the Artificial Analysis Intelligence Index at its highest thinking level, third overall, and that at the default level its cost per task is 32% lower than Gemini 4 Carbon's. Google also describes Neon directing parallel Hydrogen subagents inside Antigravity.

What it means for you. Developers now have four Gemini 4 models to choose from. A sensible approach is to start with the cheapest model that handles your task and move up only when it fails, rather than defaulting to the largest.

Google: Tasks in Gemini becomes free

What happened. On November 17, Google's Josh Woodward announced that Gemini's task mode is rolling out free of charge, running on Gemini 4 Hydrogen. You switch the Gemini app from Chat to Task, text Gemini, or forward emails to it, and it works through the job.

Who gets it. Google says the free rollout covers people in the U.S., Japan, Korea, Canada, Australia and India. A set of advanced features is starting in the U.S. first.

What it can do. Google's examples include phoning businesses to book appointments, reading bills from your email, and shopping. Purchases use the Agent Payments Protocol, with Mastercard and Visa supported and American Express listed as coming soon. Google says you can set a hard spending cap in advance, that Gemini tells businesses it is an AI when it calls, and that each task runs in a hardware-isolated virtual machine dedicated to your account. Google also says Gemini now connects with more than 60 apps.

What it means for you. This is the most direct step yet towards an assistant that acts for you. Before you connect payment cards or a password manager, decide what limits you want, and use the spending cap.

Gemini: training opt-out separated from chat history

What happened. On November 16, Google said it is updating controls in the Gemini app. A new setting lets you keep your chat history while opting out of having those chats used to train generative AI models.

The trade-off. Google also changed what happens when you turn the new History setting off: future chats will then be kept for around two months, and Google says they are used to respond to you, process feedback and protect Google, its users and the public.

Who gets it. Google says the changes will roll out globally in the Gemini app in the coming weeks.

What it means for you. If you previously turned history off only to keep your chats out of training, the new training setting is now the more precise tool. Check both settings once the update reaches your account.

Google Search: temporary threads in AI Mode

What happened. On November 17, Google added temporary threads to AI Mode. You start one with an icon in the upper-right corner of AI Mode.

How it works. Google says the session is detached from your account: answers are not shaped by your past preferences, do not affect future personalisation and are not saved to your history. Closing the thread returns you to normal search.

Who gets it. Google says it is reaching users worldwide, in mobile and desktop browsers, in all the languages AI Mode supports.

What it means for you. It is a quick way to keep a sensitive question, or a one-off topic, out of your search profile. It is not described as a full privacy mode, so account-level settings still matter.

Hugging Face: Inference Providers now route through OpenRouter

What happened. On November 16, Hugging Face said it is switching its Inference Providers router to OpenRouter's. Requests to language models now go through OpenRouter, which chooses a provider per request by weighing current price, speed and reliability, and tries another provider if the first one fails.

What changes for developers. Hugging Face says existing code keeps working. You can add a suffix to the model id to prefer the fastest or the cheapest provider, or to pin one provider. The router accepts OpenAI's Chat Completions and Responses formats and Anthropic's Messages format, so coding agents such as Claude Code can point at it.

Prices. Hugging Face says prices stay the same: the provider's rate shows up on your Hugging Face bill with no markup added, and PRO users get $2 of credits each month. Image, video and audio generation stay on Hugging Face's own Inference Providers and do not go through OpenRouter.

What it means for you. Automatic retries should mean fewer failed calls when you use open-weight models through Hugging Face, without any code changes on your side.

Snap: build Snapchat Lenses from a coding assistant

What happened. On November 18, Snap introduced Easy Lens CLI, a tool that lets AI coding assistants such as Claude Code, Codex and Cursor build, test and publish Snapchat Lenses from a plain-language description.

Who gets it. Snap says the tool is available now, free with a Snapchat account, and runs on Mac, Windows and Linux. Lenses made this way go through the same review as other Easy Lens creations, and creators in Lens+ Payouts can earn from them.

What it means for you. If you already use a coding agent, AR effects and small camera games for Snapchat are now within reach without learning Snap's own development tools.

Google DeepMind: Genie 4 opens to AI Ultra subscribers

What happened. On November 18, Google began rolling out Genie 4 in Project Genie, its tool for generating interactive worlds in the browser.

What is new. Google says you can change a world mid-session by typing a prompt, and that Genie 4 adds real-time audio. A single Genie 3 generation can now last 3 minutes instead of 60 seconds. A new agent builds worlds from a conversation, and a community gallery lets people share and remix them. All generated video and audio carries SynthID watermarks.

Who gets it. Eligible Google AI Ultra subscribers can use it now; Google AI Pro members can join a waitlist.

What it means for you. This is mostly of interest to Ultra subscribers and creators who want to experiment. Google itself notes open problems, such as camera latency and keeping scenes consistent.

Suno: Visuals adds video to AI songs

What happened. On November 18, The Verge reported that Suno launched Visuals, which generates images and video clips of up to 15 seconds from a section of a Suno song, with optional lip-sync to the vocals.

Who gets it. According to the report, Visuals works on web, iOS and Android. It uses its own Visuals credits, separate from song credits, so it costs extra.

What it means for you. It saves a trip to a separate video tool when you need a clip for social media, but budget for the additional credits.

How to use this week's changes

  • Re-check your model choice. With GPT-6.2 and Gemini 4 Neon both new, it is a good week to run your own prompts through more than one model before committing.
  • Review your privacy settings. Gemini's new training control and AI Mode's temporary threads both give you more say over what is kept.
  • Set limits before you delegate. If you try Gemini's task mode, set a spending cap and connect only the accounts it needs.
  • Treat benchmark claims as claims. Most of the figures above come from the companies themselves; independent results often arrive later and can differ.