# AI 24 Updates · 2026-11-17-2000

Published: November 17, 2026 · 20:02 CET

## In Brief

- OpenAI — GPT-6.2 Astra and GPT-6.2 Sol launch in ChatGPT, Codex and the API
- Google — Tasks in Gemini goes free in six countries, adds calls and Google Pay checkout
- Anthropic — Linked billing beta lets app customers pay for Claude usage
- Perplexity — Multiagent effort level arrives in Computer
- Mistral — Vibe desktop app runs several coding agents side by side
- Kling — Kling Image 4.0 with native 8K generation
- Suno — Visuals turns songs into video clips and lyric videos

---

## 1. OpenAI — GPT-6.2 Astra and GPT-6.2 Sol launch in ChatGPT, Codex and the API

Company: OpenAI

OpenAI released GPT-6.2 Astra, its new frontier model for the hardest projects, and GPT-6.2 Sol, an efficient model for everyday work, in ChatGPT Work, Codex and the API.

**Facts:**
- Starting today, GPT-6.2 Astra and GPT-6.2 Sol are available to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex; eligible Enterprise and Edu users get access once their workspace admin enables it. Pro $500 subscribers can use Ultrafast with both models, and Pro $200 subscribers get Ultrafast with GPT-6.2 Sol. In the API the models are gpt-6.2-astra and gpt-6.2-sol, with Ultrafast available for both at launch.
- API prices per million tokens: GPT-6.2 Astra $8 input, $0.40 cached input, $40 output (input and output 20% lower than GPT-6 Astra, cached input 60% lower); GPT-6.2 Sol keeps GPT-6.1 Sol's $2 input, $0.10 cached input, $10 output. Both models have a 1,050,000-token context window, up to 128,000 output tokens, and accept text and images; requests above 272K input tokens are billed at higher long-context rates.
- OpenAI's results: on Agents' Last Exam, GPT-6.2 Astra scores 5.2 points above GPT-6 Astra at less than a third of the cost and 5.5 points above Claude Fable 5.2; on GDPval v2 it is up 6.2 points from GPT-6 Astra with 22% fewer output tokens; on Cognition's FrontierCode 2.0, GPT-6.2 Sol comes within 0.5 points of Astra at about one-fifth the cost.
- OpenAI says both models meet the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities under its Preparedness Framework, and launch with tighter safeguards for higher-risk cybersecurity requests. GPT-6.2 Astra is OpenAI's first externally released model to meet the High threshold for AI self-improvement.
- API changes: starting with GPT-6.2, the API drops reasoning items when it detects mid-conversation edits to the conversation history or a model downgrade, and rejects reasoning effort none. A new Responses API public beta adds discounted background completion windows on GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna: 30% below Standard token rates for a two-hour window and 40% below for an eight-hour window.

**Sources:**
- [OpenAI](https://openai.com/index/introducing-gpt-6-2-astra-sol/) (official)
- [API changelog](https://developers.openai.com/api/docs/changelog) (official)
- [GPT-6.2 Astra model page](https://developers.openai.com/api/docs/models/gpt-6.2-astra) (official)
- [GPT-6.2 Sol model page](https://developers.openai.com/api/docs/models/gpt-6.2-sol) (official)
- [Codex changelog](https://developers.openai.com/codex/changelog) (official)

## 2. Google — Tasks in Gemini goes free in six countries, adds calls and Google Pay checkout

Company: Google

Tasks in Gemini is rolling out free of charge in six countries on Gemini 4 Hydrogen, with calling, Google Pay checkout, texting and email hand-off starting in the U.S.

**Facts:**
- Tasks in Gemini is rolling out free of charge to hundreds of millions of eligible users in the U.S., Japan, Korea, Canada, Australia, and India, running on Google's Gemini 4 Hydrogen model. Users flip the home-screen toggle from Chat to Task to hand off longer projects and recurring routines that run in the background.
- Starting in the U.S., Gemini can place calls to businesses on a user's behalf, including several shops at once, to compare quotes, check inventory, or book appointments; calls can be cancelled while in progress. Calling is not available in Indiana, Louisiana, Minnesota, Montana, and Nebraska.
- Gemini can now check out with Google Pay using the Agent Payments Protocol (AP2), with a user-set spending limit and store list; it asks before buying unless the user lets it proceed within the limit. Google lists Mastercard and Visa, with American Express coming soon; the feature is rolling out in English on web, Android, and iOS for U.S. users 18+ with personal accounts, and Google Wallet is now available on iOS in the U.S.
- U.S. users can also text Gemini through Google Messages or iMessage after a one-time phone verification, or forward emails and PDFs to gemini@google.com to start a task; Gemini replies in a private 1:1 thread with status updates.
- Google says each task runs in a hardware-isolated virtual machine on Google Cloud's Agent Compute Engine, dedicated to the user's account, and Gemini is designed to ask for confirmation before critical steps such as buying. Gemini connects with more than 60 apps.

**Sources:**
- [Google blog](https://blog.google/innovation-and-ai/products/gemini-app/tasks-in-gemini-free-new-features/) (official)
- [Gemini release notes](https://gemini.google/release-notes/) (official)
- [Gemini on X](https://x.com/GeminiApp/status/2122737596971503885) (official)

## 3. Anthropic — Linked billing beta lets app customers pay for Claude usage

Company: Anthropic

Anthropic launched Linked billing in beta, letting customers of apps built on the Claude API pay for their Claude usage through their own Claude organization.

**Facts:**
- Linked billing is in beta behind the user-profiles-billing-2026-10-01 beta header.
- Customers of an app built on Claude can pay for the Claude usage in that app through their own Claude organization.
- Once a customer links their organization, developers set billing_policy on each /v1/messages or /v1/chat/completions request to choose who pays.
- In the Claude Console, customers can set a spend cap for the app or unlink it. Linked billing works for customers whose organization pays Anthropic directly for API usage.

**Sources:**
- [Claude Platform release notes](https://docs.claude.com/en/release-notes/overview) (official)

## 4. Perplexity — Multiagent effort level arrives in Computer

Company: Perplexity

Perplexity is rolling out Multiagent, an effort level in Computer and teammates where a lead agent splits work across agents running different frontier models.

**Facts:**
- Multiagent is a new effort level in Perplexity Computer and teammates. A lead agent splits a task across a team of agents that use different frontier models.
- While they work, the agents share findings with one another, and the lead agent combines everything into one answer.
- Perplexity's results: on Terminal-Bench 4.1 every mixed-model team beat four copies of its lead model (by +0.008 to +0.125), as it did on WANDR (+0.015 to +0.064); the mixed team led by Gemini 4 Carbon scored highest on Terminal-Bench 4.1 at 0.727.
- Perplexity says the top mixed team on each benchmark also cost less than four copies of its lead model: $10.31 vs. about $13 per run, and $54.50 vs. $66.41. On WANDR, three of four teams finished sooner than their model alone.
- Multiagent is available to Perplexity Max and Enterprise Max subscribers on the web; mobile and desktop support is coming soon.

**Sources:**
- [Perplexity blog](https://www.perplexity.ai/hub/blog/introducing-computer-multiagent) (official)
- [Perplexity on X](https://x.com/perplexity_ai/status/2122728603909783741) (official)
- [Perplexity on X (availability)](https://x.com/perplexity_ai/status/2122738203878695042) (official)

## 5. Mistral — Vibe desktop app runs several coding agents side by side

Company: Mistral AI

Mistral released a Vibe desktop app for running several coding tasks at once and reviewing their diffs and pull requests in one window.

**Facts:**
- Mistral released Vibe desktop: users hand Vibe several tasks at once, watch the agents work side by side, then review their diffs and track pull requests in the same window.
- The desktop app is available for download from the Vibe product page. Vibe combines Mistral's chat assistant (formerly Le Chat) and its coding agent, which also runs as a CLI and in VS Code, JetBrains, and Zed via ACP.
- Mistral's claim: Vibe with its open model Le Chonk scores highest among open models on FrontierCode and Terminal-Bench.
- As a limited launch offer, developers in Europe could start Vibe Pro free for three months with a promo code; Mistral later said the codes for the day had run out.

**Sources:**
- [Mistral Vibe](https://mistral.ai/products/vibe/) (official)
- [Mistral on X](https://x.com/MistralAI/status/2122713547952324750) (official)
- [Mistral on X (offer)](https://x.com/MistralAI/status/2122713540666806397) (official)

## 6. Kling — Kling Image 4.0 with native 8K generation

Company: Kling AI

Kuaishou's Kling launched Kling Image 4.0, an image model with native 8K output, sharper text rendering, image sets and long-image editing.

**Facts:**
- Kling launched Kling Image 4.0, its new image generation model, on Kling AI.
- The model generates images natively at 8K resolution; Kling highlights natural lighting, rich detail, and intricate textures.
- Kling says it renders crisp small text and creative typography.
- It can produce coherent image sets from one prompt and supports long-image generation and editing with a full-page view and layered editing.

**Sources:**
- [Kling AI](https://klingai.com/) (official)
- [Kling on X](https://x.com/Kling_ai/status/2122707362603794601) (official)
- [Kling on X (8K)](https://x.com/Kling_ai/status/2122731313497305397) (official)

## 7. Suno — Visuals turns songs into video clips and lyric videos

Company: Suno

Suno launched Visuals on web, iOS and Android to turn songs into video clips, lyric videos and album art.

**Facts:**
- Suno Visuals is live on web, iOS, and Android. It creates images and videos for songs, including lyric videos that move with the beat, album art, and video clips from a user's photo.
- Users drag a song from their library into the Visuals tab to make clips up to 15 seconds long, then extend, edit, and re-prompt them with text; Suno says clips follow the song's structure, pick up the lyrics, and lip sync. Original characters are offered as starting points.
- Visual credits are bought separately from song credits; for a limited time, paid subscribers get their first Visual free. Suno also says Visuals is included in its MCP.
- Suno says new text and visual filters block attempts to depict specific artists, public figures, copyrighted works, and IP, and that Visuals get embedded watermarks before they can be shared off-platform.

**Sources:**
- [Suno blog](https://suno.com/blog/introducing-visuals) (official)
- [Suno on X](https://x.com/suno/status/2122717327464747307) (official)
