Anthropic Subscription Renamed and Spaceborne TPU Success

This issue focuses on Anthropic adjusting subscription tiers, Google running spaceborne TPUs in orbit, Cursor disclosing a malicious extension, and market sentiment fluctuations driven by big tech capital expenditures.

The essentials

Anthropic Renames Max and Ultra Subscriptions and Removes Some Time Limits

Anthropic renamed its Max 5x and 20x plans to Max and Ultra, keeping prices unchanged at 100 dollars and 200 dollars per month. Starting today, most verified Ultra accounts have their 5-hour usage limits removed, while new accounts retain temporary anti-abuse restrictions.

Source ↗

Google Project Suncatcher Spaceborne TPU Successfully Operates in Orbit

Sundar Pichai announced that the Project Suncatcher prototype satellite successfully booted its TPU in orbit and ran prompts, receiving the first complete response on the ground and achieving large model inference in space with data transmission back.

Source ↗

Malicious Claude Code Discovered and Removed from Cursor Marketplace

Cursor discovered that a Claude Code extension distributed through the IDE extension market was implanted with malware, affecting up to 5000 users. This version was not developed by Anthropic. Cursor removed it from the marketplace and fixed the risk.

Source ↗

In brief

Tech Giant Market Capitalization Evaporates

S&P component stocks and the Magnificent Seven saw significant pullbacks from their highs, with four entering a technical bear market, evaporating about 4 trillion dollars in total.

Alibaba Qianwen Integration

Alibaba announced that Apple Intelligence domestic users can freely choose whether to enable Qianwen integration, launching officially in December.

OpenAI Amicus Briefs Compilation

OpenAI compiled multiple amicus curiae briefs arguing that AI training should apply fair use principles to protect scientific research.

Sierra ARR Breakthrough

Sierra announced that its ARR surpassed 400 million dollars this week, with Q3 single-quarter additions exceeding 150 million dollars.

Epoch AI Evaluates GLM

Epoch AI found that GLM 5.3 can autonomously build web attacks and read computer files using unpushed security patches.

Claude Conversation Self-Evaluation

Claude immediately provides a preliminary safety evaluation of the entire conversation after answering, used to flag potential abusive behavior.

Gemini API Tools

Google announced that the Gemini API Computer Use tool is now generally available, covering two sets of APIs.