Gemini 4 Argon goes global; X Subscription and Siri AI ship

1. On Oct 13, Google made Gemini 4 Argon generally available worldwide. It is reachable on the paid Gemini API in AI Studio and Android Studio, Gemini for business, Antigravity for AI Ultra, and the Gemini app plus AI Mode in Search for AI Pro and Ultra; the API offers a 1M-token combined input/output limit and a continuation protocol, at an introductory $2 / $10 per 1M input/output tokens (cached input 95% off) until Nov 30, 2026, then $4 / $20. There is no listed free tier for the app or API, and Antigravity stays Ultra-only. Versus prior Gemini models, Google cites stronger spatial reasoning, math precision and visual coding, shortcut attempts at 2%, and nearly 2× fewer violations on an adversarial set; the Gemini macOS app also gains Antigravity's coding interface, a chat/tasks toggle that can keep working while you are away, and connections to 40+ apps, still limited to AI Pro and Ultra.

2. @SpaceXSI launched The X Subscription, one plan covering Grok, Grok Bot, Cursor and X Premium, live for new customers now and rolling out globally over the following days. Tiers are Free $0, Plus $8, Premium $30, Super $100 and Ultra $200 per month, with a shared Grok-model usage pool across Grok and Cursor that resets weekly (Premium/Super/Ultra at 4×/15×/50× Plus); team seats match individual prices and add SSO and SOC 2. Plus does not include Grok Bot; third-party models in Cursor stay on Premium and above, billed separately at API list prices; old plans are closed to new sign-ups; account unification is permanent. Premium replaces the former SuperGrok + Cursor Pro + X Premium+ stack that cost about $90/month combined, each plan includes at least 2× the coding usage of the matching Cursor plan, and team seats are cheaper per seat than old Cursor Teams with 3× coding usage; existing plans continue with optional prorated switch.

3. @Apple announced Apple TV 4K (4th generation) and HomeView with Siri AI, a new Siri on the next generation of Apple Intelligence: full-screen conversational control, personal context from a paired Apple Intelligence iPhone, systemwide app actions, far-field hands-free mics, AI camera-activity descriptions and summarized Home notifications, and on-device or Private Cloud Compute processing; pre-orders opened Oct 13 for shipping Nov 13. Siri AI is a beta (tvOS 27.2 / HomeView 27.2), English-only at launch in the US, Australia, Canada, India, New Zealand, Singapore, South Africa and the UK; HomeView is not sold in the EU; server-side features have daily usage limits, with expanded access later for a fee. A19 claims up to 50% faster CPU and 2× faster GPU than the prior Apple TV; Apple calls Siri AI an entirely new Siri, also available on Apple TV 4K from the 2nd generation on.

4. On Oct 13, @GoogleWorkspace said skills in the Gemini app began a gradual rollout due to finish by mid-November, while skills in Workspace are delayed until further notice. Gemini-app skills are open to all Workspace customers and personal accounts at all ages; skill sharing and Drive integration are still "in the coming weeks." Workspace skills remain limited to Business and Enterprise tiers and AI add-ons, ages 18+. Google had aimed to finish Rapid Release by Oct 12 and start Scheduled Release on Oct 19; both Workspace dates slipped, and skills are still positioned to replace Gems later.

Material risk reports

Anthropic published "Strengthening our defenses against distillation attacks." Fraudulent account networks (one with 20,000+ accounts) tried to extract Claude's hidden reasoning to train rivals; Zhipu distilled cyber capabilities for GLM-5.3 after Claude Fable 5's cyber safeguards degraded its attacks; Moonshot Kimi traffic was secretly forwarded to Claude, including a request Anthropic links to a PLA-affiliated user seeking CCTV analysis of an individual. Anthropic added preserved thinking on Claude Fable 5.1 and Opus 5.5 so API accounts created on or after Aug 31, 2026 cannot edit prior context; blocked earlier models from reading those models' reasoning; limited Sonnet 5.5 reasoning transcripts to the requesting organization; deployed a two-stage classifier on Sonnet 5.5, Opus 5/5.5 and Fable 5/5.1 (under 0.1% false positives); tightened limits and bans; and since Sep 24 charges again for blocked requests in these categories. Extractable reasoning fell about 10–100×; tens of thousands of accounts were banned, often within two hours; some developers had to update integrations, and a small share of legitimate requests get blocked and charged.