Daily digest: Argon lands, Grok and Cursor share a bill, and the labs argue about rules

Checked October 14, 2026. Each item links the page it came from. Nothing here is a Littlewolf benchmark.

Gemini 4 Argon is on the API and in paid Gemini apps

On October 13, Google said Gemini 4 Argon is available to developers and enterprises through the Gemini API, to Google AI Ultra subscribers in Google Antigravity, and to Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search. The introductory API price is $2 per million input tokens and $10 per million output tokens, with cached input 95% off, through November 30, 2026. After that, Google says the price is $4 and $20. The API limit it describes is 1 million unified input and output tokens.

Gemini 4 Argon launch post

Grok and Cursor are one subscription

x.ai/pricing now sells The X Subscription: Free, Plus at $8/month, Premium at $30, Super at $100, and Ultra at $200. The pool is shared across Grok and Cursor and refreshes weekly. Premium is 4x Plus, Super is 15x, Ultra is 50x. The page says third-party models inside Cursor are not included and must be purchased separately. The announcement says existing SuperGrok, Cursor, and X Premium subscribers can keep their current plan.

x.ai/pricing · Introducing The X Subscription

Anthropic says distillation harvests fell 90 to 99 percent

On October 12, Anthropic said early evidence suggests its new defenses reduced the amount of data harvested by distillers by around 90 to 99 percent. In its own test, a model trained on full Claude reasoning transcripts scored 80.4% on SWE-bench Verified. The same setup on summarized transcripts scored 71.8%. Preserved thinking blocks new API accounts, created on or after August 31, 2026, from editing Claude's prior context. Anthropic says the two-stage classifier's false positive rate is less than 0.1%, and that it has banned tens of thousands of accounts.

Strengthening our defenses against distillation attacks

Anthropic's small-model replication post

The same day, Anthropic published a research post on models copying themselves onto cheap devices. It says a 9B model got inference running on 8 of 9 hardware types tested, and that Gemma 4 26B was split across 15 Android TV boxes, about $55 each, at 0.39 tokens per second. In its simplified evals, GLM-5.2 completed end-to-end replication. The post says smaller models may match that level by the end of 2026 if they keep improving at the historical rate. The exploit step in the eval is deliberately easy: no authentication on the ADB port.

AI models may be capable of self-replicating without access to a datacenter

OpenAI and Anthropic asked UK lawmakers for binding rules

WIRED reported on October 13 that OpenAI and Anthropic told a British parliamentary committee that voluntary pledges are not enough. WIRED quotes Anthropic's Liam Booth-Smith: "The rules should be binding." It quotes OpenAI's Tom Duff Gordon saying some elements should be mandatory, starting with incident reporting, and that basing everything on voluntary commitments is not a good long-term plan. WIRED says Google DeepMind's Owen Larter answered "Focus on the standards," and that Meta's Rob Sherman pointed to existing regulators. WIRED also reports MI5 director Ken McCallum's warning about AI and bioweapons. This item is WIRED's account of the hearing, not a transcript we checked.

WIRED

GPT-6 chat rolled out on October 7

OpenAI said GPT-6 with Intelligent UI started rolling out that day to Plus, Pro, Business, and Enterprise in the Chat tab, and to Free and Go the next day. Paid chat uses GPT-6 Sol. Free and Go use GPT-6 Luna. Work and Codex models were not changed in that release. OpenAI says GPT-6 Instant starts answering web-search questions 44% sooner, on average, than GPT-5.6 Instant, in an internal evaluation.

GPT-6 and Intelligent UI for everyone

Codex can suggest the next message

OpenAI's October 9 release notes say Composer predictions are in beta for personal ChatGPT Pro users 18 and older, in the latest Codex desktop app, on local and SSH threads using GPT-6 Astra or GPT-6.1 Sol. Tab accepts the suggestion. It does not send it. During the beta, generating a prediction does not count toward Codex usage limits or consume credits. Sending the message still does.

OpenAI release notes

Ultrafast for GPT-6.1 Sol on the API

On October 8, OpenAI added Ultrafast mode for GPT-6.1 Sol in the Responses API: model gpt-6.1-sol with service_tier: "ultrafast". It is available to all API users, subject to rate limits, with global processing and US and EU data residency. The note does not print a token price.

OpenAI release notes

Claude Motion exports to Runway

Runway said on October 8 that Claude Motion is in beta on Claude Team and Enterprise, and that an animation can be exported to Runway for generated footage, scene edits, and reformatting. Claude's pricing table lists Claude Motion as unavailable on Free, Pro, and Max, and available on Team and Enterprise.

Runway · Claude pricing

Midjourney's October 7 alpha build

Midjourney's October 8 changelog, for the October 7 alpha, adds Thinking Mode on version 8.2 jobs and shared folders. Collaborators can generate into a shared folder. Viewers can browse. Sharing a folder does not change image privacy: non-stealth images can still appear on Explore and on profiles.

Alpha changelog

The plan stickers behind these stories are on AI Model & Plan Prices. That page is about model subscriptions, not the price of this site.