⚡ Quick verdict

We may earn a commission if you buy through our links, at no extra cost to you.

It was the busiest week of November so far. Between November 8 and November 14, 2026, Google shipped two Gemini 4 models in three days, Anthropic cut its flagship price, rebuilt Claude as an office suite, and marched its IPO toward a November 18 pricing date, OpenAI let its agents pay for things, and the robotics world got two open models in three days — NVIDIA's GR00T 2 and Runway's Praxis-1. Add a new decision-model category (Google and Perplexity launched competing "decision" models on consecutive days), a DNA model that reads cells one by one, and a creative-agent crew from Krea, and this is the week the AI race stopped being a model race and became a race over who does the work for you. This roundup covers the seven items that matter most to ordinary users, developers and small businesses, with a short "also this week" section for the launches that did not make the main list. None of these items repeats last week's roundup — GPT-6 Cyber, Gemini Omni 1.2 Flash, Kimi K3.1 Pro, Tencent Hy4, SWE-2 Flash, Screen Studio 4.0 and Claude Personal Finance were all covered in the November 1–7 edition.

1. Gemini 4 Hydrogen — Google's fast, cheap agentic model (November 12)

On November 12, 2026, Google introduced Gemini 4 Hydrogen, a model built for everyday coding and agentic workflows — the kind of work where a model calls tools, runs commands and iterates on its own. Google prices it flat across its entire 1-million-token context window at $1.50 per million input tokens and $7.50 per million output tokens, which makes it the cheap end of the Gemini 4 family: the flagship Carbon, which went generally available two days earlier, costs $2/$10 through November 30 and $4/$20 after that. Google says Hydrogen is about 2.5 times faster than Gemini 3.8 Flash, the model it replaces across Google's products, and it scores 57% on Terminal-Bench 4.1 — up from 18.2% for its predecessor, according to Google's own figures. Those are vendor-reported numbers, not independently verified, so treat the benchmarks as Google's claims until third parties reproduce them. Read our full story: Gemini 4 Hydrogen launch.

What matters for ordinary users is the price-to-capability ratio. Hydrogen is the model Google wants developers to reach for by default: cheap enough to run agents all day, fast enough that the agent does not keep you waiting, and available in the Gemini API and AI Studio, including a free tier — check the official Google pricing page for current limits. If you build with AI, this is the launch to trial first: swap it into an agent workflow and see whether the cheaper model finishes the same jobs. If you are a general user, the effect reaches you through the tools that adopt it — faster, cheaper agents in the apps you already use.

Try Gemini →

2. Claude Fable 5.2 + Mythos 5.2 — Anthropic cuts its flagship price 20% (November 9)

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, and priced both at $8 per million input tokens and $40 per million output tokens on the Claude API — a 20% cut from Fable 5.1 and Mythos 5.1, which were listed at $10 and $50. The two are the same underlying model with different safeguards: Fable 5.2 ships with cybersecurity and biology safeguards and is open to every API customer, while Mythos 5.2 runs without those safeguards and is limited to organizations approved through Anthropic's Cyber and Life Sciences Verification Programs. Both keep a 1-million-token context window and a 128K-token output limit, and Anthropic says output is up to 30% faster than the 5.1 generation, with an optional fast mode at $16/$80. On FrontierCode 2.0, an agentic coding benchmark, Fable 5.2 scored 49.1% at its highest effort setting against 32.4% for Fable 5.1, per Anthropic's figures. Read our full story: Claude Fable 5.2 and Mythos 5.2 launch.

The honest take is that the price cut is real but the bill is not just the price list. Anthropic's own guidance is that effort settings drive real cost per completed task more than token prices do — a cheaper model that burns more tokens on a hard task can cost more than a pricier one that finishes it cleanly. The practical move is an effort sweep: start at the default high effort, measure cost per finished task, and only then decide whether the 20% cut reaches your invoice. For Pro subscribers, the launch came with a time-limited offer: all Fable models for up to 50% of weekly usage limits through November 16, plus one limit reset.

Try Claude →

3. Claude Sheets, Whiteboards, Data — and Meetings (November 10–12)

On November 10, Anthropic turned Claude into something closer to an office suite. It launched Claude Whiteboards and Claude Sheets in beta, made Claude Data (formerly Dashboards) generally available, and added interactive connectors that open full partner apps — Adobe, Atlassian, Box and others — inside Claude. Whiteboards is a shared canvas where your team and Claude brainstorm, and Claude organizes ideas into groups, flows and timelines. Sheets is a spreadsheet you and Claude edit in the same cells, with tabs, formulas, formatting and drop-downs, exporting to .xlsx or CSV, capped at 100,000 cells per sheet. Data turns company information into dashboards, data stories and narrated videos from warehouses and CRMs such as Amazon Redshift, BigQuery and ClickHouse. All three are included in every paid Claude plan at no extra cost; the free plan does not get them. Then, on November 12, Claude Meetings started rolling out in beta on Team and Enterprise plans, joining Google Meet calls to take notes and produce summaries. Read our full story: Claude Sheets, Whiteboards and Data launch.

Two caveats belong next to the excitement. First, the pieces do not share live data with each other yet — a board, a sheet and a dashboard are three separate surfaces, not one workspace. Second, the defaults differ by plan: on Team plans all three are on by default, while on Enterprise plans Whiteboards and Sheets start off and Claude Data switches on by default on November 19, 2026. Sheets also does not yet support HIPAA, customer-managed encryption keys or zero data retention. If you already pay for Claude, the honest test is a real team task — plan a launch on a whiteboard, track it in a sheet, report it from Data — and see whether the suite replaces a tab you currently keep open. If you do not use Claude, this is not yet a reason to switch; it is a sign that the assistants are competing on the whole workday now, not just the chat box.

Try Claude →

4. ChatGPT Wallet — OpenAI's agents can now pay (November 13)

On November 13, 2026, OpenAI launched ChatGPT Wallet, which it calls the first wallet that lets agents pay without access to users' actual card details and that requires a security check for every purchase. The wallet lives inside dots, the always-on agents OpenAI introduced in September, and it was built in partnership with American Express, Mastercard and Visa, with Forter, Riskified, Sardine and Signifyd on fraud detection and Adyen, Checkout.com and Stripe on the payment side. OpenAI takes a 0% cut on purchases at launch, per CNBC's reporting. The rollout is narrow: it starts with ChatGPT Pro subscribers in the US who are 18 or older with eligible US-issued cards, and it only works inside dot conversations — not in Chat, Work, or standalone Codex threads. Read our full story: OpenAI ChatGPT Wallet launch.

The wallet solves payment. It does not solve permission, and permission is the whole ballgame for agent commerce. Two days before the launch, United Airlines rewrote its Contract of Carriage to restrict unauthorized third-party AI agents from buying tickets — a reminder that merchants decide whether an agent may buy, whatever the wallet allows. The honest take: for ordinary users this changes nothing today unless you are a US Pro subscriber with a dot set up; for everyone else it is the clearest signal yet that the AI companies see checkout, not chat, as the next battleground. Watch which merchants accept agent purchases and which block them — that list, not the wallet itself, will decide whether this matters.

Learn more at OpenAI →

5. Krea Bots — a creative crew of nine AI agents (November 12)

Krea opened the beta for Krea Bots on November 12, a system of nine named AI agents that handle specific creative tasks, retain approved preferences across projects, and pass work through a shared chat. You message a bot like a colleague and get finished work back: a billboard for a product launch, a campaign's visual assets, a video edit. Krea says all of its recent creative work — including the Bots launch video itself — was produced by these bots, and it estimates 43 hours of bot work and seven minutes of human input for its example product launch, though it provides no benchmarking methodology for that figure. Accepted teams receive $1,000 in credits to test the full feature set; bots-specific pricing and a general availability date remain unpublished. Our full Krea review covers the platform in depth, including the Bots beta.

Why it matters in a week of giant models: Krea Bots is the most concrete version yet of the "agent does the project" idea — not a chatbot that answers questions about a video, but a crew that makes the video, remembers your taste, and hands it back. The honest caveat is that $1,000 in credits is only as valuable as Krea's metering: what the credits buy depends on how Krea charges for model calls, compute, storage and connected-tool actions, and none of that is published yet. If you run a creative team, the beta is worth applying for — the credits cost nothing to try. If you are a solo creator, read our Krea review first: the core platform's free tier (100 compute units a day) is the place to start, and Bots is a team product.

Join the Krea Bots beta →

6. Grok Bot 1.0 — shared AI teammates in Slack and Teams (November 11)

Grok Bot left beta on November 11, 2026, as Grok Bot 1.0, and the change is not the model behind it — it is who the bot works for. A Bot can now be shared with a whole team, added to a Slack workspace or a Microsoft Teams team, and invited into a Google Meet, Zoom or Teams call, where it answers in the meeting chat and sends a summary afterwards. Bots can also offer to take on work without being asked, and teammates can teach a Bot things it has learned — with the owner approving what the whole team gets, under a "Review learnings" step. Team Bots need a Team or Enterprise plan, and meetings need Team, Enterprise or Ultra. The launch post is dated November 2 in its metadata while the public announcement came on November 11, so treat 1.0 as the point where these features came out of beta together, not as a single day when everything appeared.

The honest comparison is with the week's other agent launches: Claude is adding office tools, Krea is adding a creative crew, and Grok Bot is adding a teammate that lives in the channels your team already uses. The pitch is the same in all three cases — the agent moves from answering one person to working inside a team — and the test is the same too: does the bot actually save a meeting, or does it just add another participant that needs managing? Grok Bot's answer depends on your stack: if your team lives in Slack and Meet, the shared bot is a genuine workflow change; if it does not, this is a version number, not a reason to switch.

Learn more at xAI →

7. Anthropic's IPO roadshow — pricing expected November 18

Anthropic's investor roadshow, which launched November 10 with an amended S-1 filing, is the biggest capital-markets story in AI this year. The company is offering 825 million Class A shares at $70 to $78 each, for a Nasdaq listing under the ticker ANT and a fully diluted valuation of up to $1.4 trillion, raising as much as $64.4 billion. Bloomberg reported on November 12 that orders reached about six times the shares available, and the Financial Times reported bankers expect to price above the $70–$78 range, lifting the valuation target to $1.5 trillion or more — reported expectations, not filed terms, and no new price range has been submitted to the SEC. On November 13, CEO Dario Amodei joined JPMorgan chief executive Jamie Dimon at a livestreamed investor event. Pricing is expected November 18, with trading planned for November 19. Read our full story: Anthropic IPO roadshow.

Why it belongs in a tools roundup: the IPO is the moment the AI labs stop being venture-funded research projects and start being public companies with quarterly earnings, and that changes what they build. Anthropic's filing highlights $17 billion in third-quarter revenue and a $23.3 billion net loss, including a $21.9 billion non-cash valuation-related charge, per its financial presentation. The honest take for users is simpler: a public Anthropic will be judged on revenue growth, which means more products like Claude Sheets and Claude Data — tools that sell to teams, not just chatbots that sell to individuals. The November 18 pricing date is the one to watch; the final price is set the evening before trading starts.

Learn more at Anthropic →

Also this week: seven more launches worth knowing

Seven more items from November 8–14 did not make the main list but are worth a line each. Gemini 4 Carbon went generally available on November 10 — Google's top-tier model, announced October 28 and first given only to cybersecurity defenders through the Fairwind Program, now open to all paying developers and enterprises at $2 per million input tokens and $10 per million output tokens through November 30, then $4/$20 from December 1. Gemini Decision 1 arrived November 11 with the Gemini Decisions API: a model that answers up to 1,000 closed questions in a single pass and returns calibrated probabilities instead of text, at $0.10 per million input tokens with zero output tokens billed. Perplexity Decider v1.3 followed on November 12 with the same idea — open Apache-licensed weights, a new Locate feature that extracts exact text spans, a reported 91.6% mean accuracy across 63 tasks, and free access on OpenRouter until November 19. AlphaGenome 2, from Google DeepMind on November 11, predicts how DNA variants affect gene regulation in individual cell types — recovering 30% of causal variants on one immune benchmark against 16% for its predecessor — free for non-commercial research through the AlphaGenome API, and not validated for clinical use. Runway Praxis-1, open weights on November 11, is a 27-billion-parameter "world action model" that turns video pretraining into robot control, scoring 46.2% on NVIDIA's RoboLab-120 benchmark by Runway's own numbers, under Apache 2.0. NVIDIA GR00T 2, unveiled November 9 at the Conference on Robot Learning, is the other robot model of the week: it learns dexterous skills from human video, reaching 95% success across 50 tasks with five demonstrations each, per NVIDIA's figures — but the weights are not downloadable yet, with code listed for December. And Fish Audio Live 3 with Fish Agent, announced November 10, is a full-duplex voice model that keeps listening while it speaks and can start a short reaction in as little as 186 milliseconds, with a no-code agent builder that Fish says can deploy a basic phone agent in about 30 seconds, at a promotional $0.035 per minute through December 31, 2026.

New in the Aitopkit directory this week

Thirteen tools joined the Aitopkit directory this week: Grok Bot, Krea Bots, Fish Audio, CALL-E, Nod, ElevenAI, MiniMax H3-Image, Memorias, Agent Clan, Glasshand, Cardboard AI, Google Vids and Google Antigravity. The pattern is hard to miss — agents and video dominate the new listings, with three agent platforms (Grok Bot, Krea Bots, Agent Clan), three video and image tools (Fish Audio's voice agents, MiniMax H3-Image, Google Vids) and a new agentic development platform in Google Antigravity. Browse the directory categories to try them, and see our best AI video generators guide for where the video newcomers fit.

The pattern of the week: the agent does the work now

Step back from the individual launches and this week has a shape, and it is the same shape in every lab's announcement. Three things happened at once. First, the models got cheaper and faster at the same time: Hydrogen at $1.50/$7.50, Fable 5.2 down 20%, Carbon's intro price held at $2/$10 — the frontier is no longer priced like a luxury, and the labs are competing on cost per finished task. Second, the models stopped generating text and started making decisions: Gemini Decision 1 and Perplexity Decider v1.3, launched a day apart, both return probabilities instead of prose, because agents need answers they can branch on, not paragraphs they have to parse. Third, the products around the models became the product: Claude's office suite, Krea's creative crew, Grok Bot's shared teammate, ChatGPT's wallet — in every case, the value is not the model, it is the work the model does inside a team, a project or a purchase.

That is the story of AI in late 2026: the model is becoming infrastructure, and the race is over the workflow on top of it. The launches to watch next are not the next model announcements, but the first evidence of whether these agents actually save time — the first independent tests of Hydrogen's speed claims, the first teams that run a real project through Krea Bots, the first merchants that accept or block an agent purchase. Next week's roundup will cover what these launches actually turned out to be once users got their hands on them, which is when the interesting part starts.

What to do with this week

If you take one action from this roundup, try Gemini 4 Hydrogen in AI Studio — it costs nothing on the free tier and it is the model Google wants every agent to run on. If you build on Anthropic, run the effort sweep on Fable 5.2 before you celebrate the 20% cut: measure cost per finished task, not cost per token. If you run a creative team, apply for the Krea Bots beta while the $1,000 credits are on offer — and read our Krea review first so you know what the platform does without the bots. If you use Perplexity's Decider, try it free on OpenRouter before November 19, when the free window ends. And if you buy AI tools at all, note the pattern of the week: the cheapest model that finishes the job is the right model, and the tool that does the work for you is worth more than the tool that only answers questions.

Frequently asked questions

What was the biggest AI launch this week?

Google's Gemini 4 Hydrogen, launched November 12, was the week's biggest model launch: a fast, cheap agentic model at $1.50 per million input tokens and $7.50 per million output tokens, scoring 57% on Terminal-Bench 4.1 according to Google. Anthropic's IPO roadshow, with pricing expected November 18, was the week's biggest business story. Both are covered in detail above.

Which of this week's AI launches can I actually use for free?

Gemini 4 Hydrogen is available in the Gemini API and AI Studio, including a free tier — check the official Google pricing page for current limits. Perplexity's Decider v1.3 is free on OpenRouter until November 19, 2026, and its weights are Apache-licensed on Hugging Face. Runway's Praxis-1 is open-weight under Apache 2.0, though running it needs a powerful GPU. Krea Bots is in beta with $1,000 in credits for accepted teams.

What is the difference between Claude Fable 5.2 and Mythos 5.2?

They are the same underlying model with different safeguards. Fable 5.2 has cybersecurity and biology safeguards and is open to every API customer at $8 per million input tokens and $40 per million output tokens — a 20% cut from the 5.1 generation. Mythos 5.2 is the same model without those safeguards, limited to organizations approved through Anthropic's Cyber and Life Sciences Verification Programs.

Can ChatGPT Wallet actually buy things for me?

Yes, within narrow limits. ChatGPT Wallet, launched November 13, lets OpenAI's dot agents complete purchases without seeing your actual card details, with a security check on every purchase. At launch it is only for ChatGPT Pro subscribers in the US who are 18 or older with eligible US-issued cards, and it only works inside dot conversations — not in regular Chat, Work, or standalone Codex threads.

Where can I follow next week's AI launches?

Aitopkit publishes a weekly roundup every Saturday covering the week's verified launches, plus daily AI news. Bookmark the articles index to catch each week's edition as it appears.

AT
About Aitopkit

We research AI tools in depth before recommending them — official pricing, public docs, and real user reports. If you buy through our links we may earn a commission at no extra cost to you. Read our disclosure.