Home » Tech News November 18, 2026: GPT-6.2 Lands, Microsoft Bets on Cheap AI, and Gemini Starts Doing Your Errands
Posted in

Tech News November 18, 2026: GPT-6.2 Lands, Microsoft Bets on Cheap AI, and Gemini Starts Doing Your Errands

Abstract illustration of a glowing AI network core connected to a robot hand, laptop, smartphone and shield over a city skyline at dusk

Tech News November 18, 2026: GPT-6.2 Lands, Microsoft Bets on Cheap AI, and Gemini Starts Doing Your Errands

Tuesday was the busiest model-launch day in weeks, and the theme was unmistakable: the AI race is now being fought on price and usefulness, not just raw capability. OpenAI shipped GPT-6.2 on the same day Microsoft unveiled seven cheaper in-house models at Ignite, while Google made its do-it-for-you Gemini agent free for hundreds of millions of people. Add in a major breach at an AI vendor and the biggest IPO in tech history pricing tonight, and it was a lot to keep up with. Here’s what mattered.

Key Takeaways

  • OpenAI launched GPT-6.2 Astra and Sol, with Astra 20% cheaper than its predecessor — though independent tests put the real savings well below OpenAI’s headline claim.
  • Microsoft unveiled seven MAI models at Ignite, including coding and security models it says match frontier rivals at a fraction of the cost.
  • Google made Gemini Tasks free in six countries, turning Gemini from a chatbot into an agent that calls businesses, books appointments, and pays bills.
  • A breach at AI vendor Fireworks exposed the personal data of nearly 150,000 Uber drivers, plus data at Cursor and Notion.
  • Klarna launched an AI concierge that shops, books, and chases returns over WhatsApp and iMessage.
  • Anthropic prices its IPO tonight, with trading set to start Thursday under the ticker ANT.

OpenAI Ships GPT-6.2 Astra and Sol

According to OpenAI, GPT-6.2 Astra and GPT-6.2 Sol launched on November 17 in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, plus the API. Astra is the new flagship for the hardest professional work; Sol is the cheaper everyday model. Neither is coming to regular ChatGPT chat — this rollout targets paying work users and developers.

The headline change is price: Astra now costs $8 per million input tokens and $40 per million output tokens, down 20% from GPT-6 Astra, while Sol holds at $2 and $10. OpenAI says Astra delivers new highs at less than a third of GPT-6 Astra’s cost per task, but independent checks are more modest — Artificial Analysis measured a 36% cost-per-task drop, Vals AI about 50%. Real savings, just not the headline number. Worth watching on safety: Astra is the first OpenAI model deployed externally rated “High” for AI self-improvement risk, and both models are rated “Critical” for cybersecurity capability.

Microsoft’s Ignite Answer: Seven Cheaper Models of Its Own

According to Microsoft, CEO Satya Nadella opened Ignite 2026 in San Francisco by unveiling seven in-house MAI models spanning coding, security, image editing, voice, transcription, and routing — priced to undercut the frontier labs. The headliners are MAI Code 2 and MAI Code 2 Flash, agentic coding models now generally available in GitHub Copilot and Foundry. Microsoft says Code 2 delivers 98% of GPT-6.1 Sol’s coding quality at about 40% of the price, and 96% of Claude Opus 5.5’s at roughly 5% of the cost. MAI Cyber 2 Flash reportedly matches GPT-6 Astra on a cybersecurity benchmark at under 2% of the cost per task, though it’s only available to verified defenders inside Microsoft’s Codename MDASH program. The honest caveat: every performance and price figure here is vendor-reported, and some cost comparisons rest on promotional pricing. Still, the direction is clear — Microsoft wants to own the cheap end of enterprise AI instead of renting the frontier.

Google Makes Gemini Tasks Free

According to Google, Gemini Tasks is rolling out free starting November 17 to hundreds of millions of people in the US, Japan, Korea, Canada, Australia, and India, powered by the new Gemini 4 Hydrogen model. Tasks is the new name for Gemini Spark, the agent Google showed at I/O in May that was previously limited to paid Pro and Ultra subscribers. Flip the Gemini home-screen toggle from Chat to Task, describe what you want done, and Gemini works on it in the background.

Tasks can find local businesses, call several at once to book appointments, add the result to your calendar, and pay and check out through Google Pay using a new Agent Payments Protocol, with a spending cap you set. It can also log in to sites for you without sharing your passwords, and connects to more than 60 apps, each task running in its own hardware-isolated virtual machine. Two catches: the fanciest features start in the US only, with calling unavailable in five states, and Google is upfront that tasks can fail or miss your intent. Agents that spend your money are a trust exercise, and this is the biggest one Google has run yet.

A Breach at AI Vendor Fireworks Exposes Nearly 150,000 Uber Drivers

Not all the agent news was cheerful. According to BleepingComputer, Uber is notifying nearly 150,000 drivers that attackers stole their personal data after breaching AI infrastructure provider Fireworks in early October. A filing with Maine’s attorney general puts the count at 149,346 drivers, with names, driver’s license numbers, dates of birth, and addresses exposed — and images of driver’s licenses, according to some reports.

Uber says it learned of the incident on October 9, confirmed by October 13 that attackers had used compromised service-provider credentials to reach data Uber had shared with the vendor, and then cut its data connections. Fireworks attributed the attack to a financially motivated group linked to large-language-model hijacking attacks. The blast radius goes beyond Uber: Cursor named Fireworks as the origin of its own security incident, and Notion confirmed some AI chats were exposed. Uber is offering free credit monitoring and says it has found no evidence so far that the stolen data has been misused. The lesson is an old one with a new wrapper — your data is only as safe as your vendor’s vendors.

Klarna’s AI Assistant Wants to Run Your Errands on WhatsApp

The agentic wave isn’t just for enterprises. According to Klarna, the payments company announced its Klarna AI-assistant on November 17, a concierge that handles shopping, returns, bookings, and everyday admin — and it lives where you already chat: WhatsApp and iMessage, as well as inside the Klarna app.

It finds options and helps you choose, books anything from a table for two to a babysitter, keeps track of deliveries and reservations you’re waiting on, drafts email replies for you to approve, and files receipts. On the shopping side it searches, compares prices, places the order — and in Klarna’s framing, you don’t pay until it arrives — then handles the return if you change your mind. It’s live now for a small group of early users, with a waitlist opening this week. The launch landed alongside Klarna’s third-quarter results: revenue up 12% to $1 billion, with a $65 million net loss. Between Klarna and Google, November 17 may go down as the day AI assistants stopped answering questions and started doing chores.

Anthropic Prices Its IPO Tonight

One more thing to watch after the market closes: Anthropic is due to price its IPO this evening, with shares expected to begin trading Thursday, November 19, on the Nasdaq under the ticker ANT. The offering is officially marketed at $70–$78 a share for 825 million Class A shares, but the Wall Street Journal and CNBC report it could price above $80 on demand running roughly six to seven times the shares available. At the top of the filed range, the deal would raise about $64 billion and value the Claude maker at roughly $1.4 trillion fully diluted — the largest tech IPO in history. Yesterday’s roundup flagged this as IPO week; tonight is when the number gets set.

The Takeaway

Three threads ran through Tuesday. First, the frontier is getting cheaper: GPT-6.2, Microsoft’s MAI family, and Gemini 4 Hydrogen are all fighting on cost per task, not just benchmark scores. Second, agents are now doing real things — calling businesses, spending money, booking tables — which makes the trust and safety questions from yesterday’s OpenAI lawsuit story feel a lot more immediate. And third, the AI supply chain is the weak link: the Fireworks breach hit Uber, Cursor, and Notion through one shared vendor. The models keep getting better and cheaper. The infrastructure around them still has catching up to do.

Check back tomorrow for the next roundup — including how Anthropic’s debut actually trades.

FAQ

What is GPT-6.2 Astra?

GPT-6.2 Astra is OpenAI’s newest flagship AI model, launched November 17, 2026, for complex professional work in ChatGPT Work, Codex, and the API. It costs $8 per million input tokens and $40 per million output tokens, 20% less than GPT-6 Astra.

What is GPT-6.2 Sol?

GPT-6.2 Sol is OpenAI’s cheaper, everyday model, launched alongside Astra on November 17, 2026. It costs $2 per million input tokens and $10 per million output tokens, and is aimed at routine coding, writing, and multi-step tasks.

What are Microsoft’s MAI models?

MAI models are Microsoft’s in-house AI models, unveiled at Ignite 2026 on November 17. The lineup includes MAI Code 2 and Code 2 Flash for coding, MAI Cyber 2 Flash for security, plus image, voice, transcription, and routing models, priced well below frontier rivals.

What is Gemini Tasks?

Gemini Tasks is Google’s AI agent in the Gemini app that carries out multi-step jobs — calling businesses, booking appointments, paying bills, and shopping — instead of just answering questions. It rolled out free on November 17, 2026, in six countries, running on the Gemini 4 Hydrogen model.

Sources

Leave a Reply

Your email address will not be published. Required fields are marked *