Changelog

#035: Gemini 3.8, Grok 4.6 & Channel Filters

Invent #035: Gemini Flash 3.8, Grok 4.6, and Muse Spark 1.3 in your model picker, plus Inbox channel filters and WhatsApp template alerts.

Sep 4, 2026

#035: Gemini 3.8, Grok 4.6 & Channel Filters
Blog/Changelog/#035: Gemini 3.8, Grok 4.6 & Channel Filters

September 4, 2026

TL;DR:

  • Three new models in the picker: Gemini Flash 3.8, Grok 4.6, and Muse Spark 1.3 are ready to power your assistants.
  • Voice notes and videos, understood: Gemini Flash 3.8 takes audio and video, and Muse Spark 1.3 takes video, so your assistant can work with what customers actually send.
  • WhatsApp template alerts: an email the moment a template is approved, rejected, or changes category.
  • Channel filters: filter the Inbox by channel, and see each contact's channel right on their panel.

It's Friday and summer isn't over at Invent. Here's what shipped this week: three new brains for your assistants, more visibility for you, and a beach.

Three new models: Gemini Flash 3.8, Grok 4.6, Muse Spark 1.3

Your assistant's model picker just grew by three. And this is the part we care about most: you control the model. You decide which one runs each assistant, and you can switch anytime, no migration needed. More on why the choice matters: using different AI models for customer support.

Gemini Flash 3.8 is Google's smartest Flash model, with a 1M-token context window and tunable thinking levels. It takes text, images, documents, audio, and video, which makes it the one to pick when your customers send voice notes.

The Gemini Flash 3.8 model card in Invent's model picker: 1M context, three speed bolts, two of three intelligence bulbs, and support for text, images, documents, audio, and video

Gemini Flash 3.8: fast, 1M context, and it takes everything a customer can send.

Grok 4.6 is xAI's flagship, with configurable reasoning and a 500K-token context window. It scores top marks on intelligence in the picker and takes text and images.

The Grok 4.6 model card in Invent's model picker: 500K context, three speed bolts, three intelligence bulbs, and support for text and images

Grok 4.6: top marks on intelligence, text and images in.

Muse Spark 1.3 is Meta's frontier model for long-running agentic work, with a 1M-token context window. It takes text, images, PDFs, and video.

The Muse Spark 1.3 model card in Invent's model picker: 1M context, and support for text, images, documents, and video

Muse Spark 1.3: built for long-running agentic work, with a 1M context window.

Note: Gemini Flash 3.8 understands voice notes and videos, and Muse Spark 1.3 understands videos. On WhatsApp, customers barely send anything else. An assistant on Gemini Flash 3.8 gets the voice note, the video of the leaky faucet, and the PDF invoice without anyone transcribing a thing.

WhatsApp template status notifications

Now, whenever a template is approved, rejected, or changes category, everyone who manages your org's channels gets an email the moment it happens.

Two Invent notification emails: one with a green Approved tag confirming the WhatsApp template order_confirmation is ready to send, and one with an amber Category changed tag showing the template order_ready moved from Utility to Marketing

Approved, rejected, or recategorized: template news lands in your email now.

New to templates? Start with our WhatsApp templates guide.

Channel filters in the Inbox

In the Inbox you can now filter your chats by channel, alongside assistant, assigned team, segment, and CSAT. A couple of WhatsApp numbers, a Telegram bot, Messenger, Slack: see exactly the slice you need.

The Invent Inbox filter panel with the Channel dropdown open, listing All Channels, a Slack workspace, and two WhatsApp numbers, each paired with the assistant that runs it

One org, many channels: filter the Inbox down to the one you need.

Each contact's panel now shows their channel too, so you always know which number or bot a conversation lives on.

A beautiful beach

Open useinvent.com and you'll land on an upgraded beach. Turquoise water, a palm tree, coconuts drifting by, and a little control on the right that changes the view. Go take a look, and if you like it, save it as your wallpaper.

The Invent homepage seen from above: turquoise water, a sandy shore with a palm tree and coconuts, the hero card floating over the sea, and a view control on the right edge

The Invent homepage, now a beach. Wallpaper-worthy.

WhatsApp updates

Starting October 1, Meta (the company behind WhatsApp) will charge a small fee for every reply your business sends on WhatsApp. To keep sending, your WhatsApp Business Account needs a payment method on file by September 30. If there's no card, your replies stop going out on October 1. It takes two minutes, and our step-by-step guide shows exactly where to click.

And some good news: the first 1,000 replies each month are free. Meta only charges from reply number 1,001, and the counter resets every month for each business phone number. So if your business sends fewer than 1,000 WhatsApp replies a month, you pay nothing. This is straight from Meta's WhatsApp pricing page.

Two details worth knowing:

  • It's per phone number. Each WhatsApp business number gets its own 1,000. Two numbers, two allowances.
  • It covers replies only. A reply is a message you send inside the 24-hour window after a customer writes to you. Template messages you send on your own initiative, like promotions, order updates, or reminders, are paid from the first one.

This is Meta's fee for using WhatsApp, and Meta bills you for it directly. It has nothing to do with your Invent plan.

Why it matters

Models decide what your assistant can understand, and two of this week's three hear voice notes and watch videos. The template alerts and channel filters are about staying in control as you grow: more channels, more automation, and you still see everything the moment it changes.

With love,
The Invent Team

Start Building Your Assistant For Free

No credit card required.