Tuesday, 13 October 2026

Google has released Gemini 4 Argon, its new frontier model, to developers through the paid Gemini API and to paying subscribers in its apps [1]. In the Gemini app and AI Mode in Search, Argon is for Google AI Pro and AI Ultra subscribers; in Google Antigravity it is for AI Ultra subscribers only [1]. According to [3], AI Pro costs $20 a month and AI Ultra starts at $99 a month, and Google has declined to say when, or if, Argon will reach Antigravity on other tiers. According to [4], it is not clear whether Argon will reach Plus subscribers or free users. Google's API pricing page says Argon is not available on the free tier [2].

API pricing starts low and doubles. Argon costs $2 per million input tokens and $10 per million output tokens through November 30, 2026, then $4 and $20 from December 1 [2]. The output price includes thinking tokens, and cached input costs $0.10 per million, rising to $0.20; only implicit caching is supported [2]. Google says the model has a 1M unified input-output token limit through the Gemini API [1].

The performance claims are Google's own. The company says Argon ranks first on the Vals Index, scores 87% on the ERQA embodied-reasoning benchmark, and cuts attack success on Gray Swan's indirect prompt injection benchmark to 0.7% across 15 adaptive attempts [1]. According to [4], Google says Argon beats GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 on a range of benchmarks. Google also says it trained Argon to pause for human confirmation before irreversible steps, and runs an automated monitor that reads the agent's reasoning to block unintended workarounds [1].

If you build on the API, price your workloads at the December rate, not the launch rate.

Character.AI has filed for Chapter 11 bankruptcy protection in the US to sell the company and its assets, the company said in a press release and blog post published Monday [1]. In its filing it listed estimated liabilities between $50 million and $100 million [1]; according to [2], the court document puts both assets and liabilities in that range. Disney would act as the lead bidder for Character.AI's assets, setting a baseline for competing offers [1]. Under the proposal, Disney would pay $20 million for certain undisclosed assets, provide transition support "using former Company employees it has hired," and receive a nonexclusive license to unspecified intellectual property [1]. Disney recently hired Character.AI's former CEO Karandeep Anand as chief technology officer, according to [1].

The service keeps running for now. Character.AI says it will continue operating, including paid subscriptions, during the sale, which it expects to finish in late November or early December [1][2]. It also warns: "If such a bid is unavailable, however, the platform will be wound down" [2]. According to [2], there is not yet a bid for the assets or a buyer.

The history explains the filing. In August 2024, Google struck a $2.7 billion deal to license the startup's models, and co-founders Noam Shazeer and Daniel De Freitas returned to Google [2]. The company was later sued following the suicide of a 14-year-old Florida boy whose mother says he became obsessed with a chatbot on the platform; it settled some lawsuits, added safety features and banned minors [2]. According to [2], high engagement cuts both ways for a consumer AI app, because AI costs rise the more people use it.

If you pay for Character.AI, or built characters there that you want to keep, save what you can now. Then watch who buys the assets and what they say about user data.

Google Play now lets people run apps they built with AI on their own Android phones without publishing them to the store, Google announced on Tuesday [1]. The apps stay private and run under the user's standard Google account [1]. According to [2], you send an app to the Play Store directly from AI Studio, with no APK files or developer console, and it lands in a "Personal Library" in the Play Store app that only you can see by default. Private apps are not reviewed by Google [2]. The feature starts rolling out today in the US and select other markets [2].

Sharing comes later. Google says users will be able to share their apps with a network of trusted friends in the months ahead, without going public [1]. According to [2], Google has not yet decided how many people count as a "small group", and this kind of sharing does not need a developer account. Publishing to everyone will go through an onboarding process for a developer profile and the same review guidelines other developers face, with the private-to-public option arriving in the coming months [2].

Sam Bright, Google's VP and GM for the Google Play and Developer Ecosystem, said "the future of apps on Google Play is personal, secure, and built by everyone" [1]. The same launch expands the Ask Play assistant to compare similar apps, and adds Game Finder, an AI game recommender that starts in the US [1][2].

For a solo operator, this removes the hardest step between a vibe-coded tool and your own phone. It also removes Google's review. An app built by someone who is not a developer still needs someone to check what data it reads and where it sends it, especially once you share it.

SpaceXSI has merged X Premium, Grok and Cursor into one paid plan called the X Subscription [1][2]. According to [2], the company acquired the coding startup Cursor in August, and was known as SpaceXAI until Elon Musk said he would rename it. There are five plans: free, Plus at $8 a month, Premium at $30, Super at $100 and Ultra at $200, with AI usage caps rising by tier [1]. The free plan offers only limited chat and agent usage, and the latest Grok models need Premium or higher [1].

According to [2], Plus includes Grok Chat, Voice, Imagine and Build, the Grok API, Cursor for coding, and X Premium features such as a checkmark and fewer ads. Premium adds Grok Bot, the latest Grok models, image and video generation and faster responses, plus Cursor code hosting, cloud agents and automations, and ad-free X [2]. Super adds priority access to Grok models at peak times, and Ultra gives the earliest access to products and models and 20 times the usage capacity of Plus [2]. Premium, Super and Ultra are also sold to teams, and enterprise customers can get custom plans [2].

Existing users can switch from their subscription settings on X or keep their current plan, and the upgrade option rolls out gradually [1]. Neither report gives the Cursor usage limits inside each tier, so how much coding the $8 plan buys is not public.

If you pay for Cursor separately, compare your current plan with the bundle before you switch. Check the coding limits on each tier, and whether your team's billing can sit under an X account.

American Express has published Intent Intelligence, a framework for capturing and authenticating what a customer wants to buy before an AI agent completes the purchase, and then carrying that intent through the rest of the transaction [1]. The company says its approach differs from current ones by giving the card issuer the customer's authorised intent earlier, before the agent buys [1]. It is available through the Amex Agentic Commerce Experiences (ACE) Developer Kit [1].

It works in steps. A Create Intent API captures the customer's instructions and constraints, such as the merchant, a maximum budget and other preferences; the customer then authenticates and signs off on the intent [1]. An Intent ID and a Proof of Intent Token follow, and the token is used to obtain a payment credential that the agent uses at checkout instead of the real card number [1][2]. Amex says the ID and token can also help with fraud prevention and dispute resolution, according to [2].

Hannah Bozian, Amex's VP of Agentic Partnerships and Strategy, gave an example: a customer asks for white Nike sneakers in size 9 and the agent tries to buy Adidas; Amex could stop that transaction [2]. Amex says it has not yet seen an increase in disputes from agentic commerce [2]. It cites its own Trendex survey, in which 66% of surveyed US consumers said they are likely to use AI agents to research the best price over the next 12 months [2]. Amex also says it is engaging with EMVCo and supporting the x402 Foundation's work on payment standards [1].

If you let an agent buy for you, the idea works at any scale: write the limits down before the agent acts, and use a payment method that enforces them.

Pluto Security has come out of stealth with $42 million in total funding to monitor and control the AI tools, agents and apps that employees use and build [1][2]. The money spans an $8.5 million seed round led by TLV Partners and a $34 million Series A led by Jump Capital, with Skinos, Ansa Capital and angel investors taking part, according to [1]. The company plans to spend it on product, security research and US growth, and has 40 employees in the US and Israel [1].

The pitch is that people outside engineering now build software. CEO Shahar Bahat told TechCrunch that about 60% of the risks Pluto detects across its customers are in tools vibe-coded by non-developers, with finance teams the biggest source, followed by marketing and legal [1]. She says a newly announced AI tool can show up inside a company within two hours of its release [1]. "There was not one POC in which we entered and didn't find malicious stuff," she said [1].

The software watches the AI tools employees run and the skills, extensions and MCP servers they rely on, and can step in while an agent works, for example by stopping an action that seeks more permissions than the user intended [1]. Its customers include Oscar Health, Workato, Sixt and AlphaSense [1][2]. The reports differ on how many customers there are: TechCrunch says more than 30 enterprise customers and more than $2 million in annual recurring revenue [1], while [2] says Pluto announced more than 40.

You do not need a vendor to start. List the skills, extensions and MCP servers your team has installed, and remove the ones nobody can explain.