Anthropic files for a Nasdaq IPO and calls Claude Code its defining product

Anthropic filed with the SEC to go public and intends to list on Nasdaq. Share count and price aren't set. [The filing](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm) shows revenue of $4.6 billion in 2025 and $16.3 billion in the first half of 2026, up from $1.2 billion in the same half of 2025, on an operating loss of $693 million for that half. Customers spending $100,000 or more a year grew from more than 950 to more than 11,900 in a year. Claude Code gets its own section. Anthropic calls it "the product we believe has most concretely defined what autonomous AI work looks like," and gives one usage number: in the September quarter, the heaviest 5% of Claude Code users on Enterprise plans spent about 20 times the average non-user. The risk factors say agents like Claude Code and Cowork "will require broad permissions and deep integrations" with customers' internal tools.

Kimi K3.1 Pro rivals Claude at coding tasks, Moonshot says, but can't be downloaded yet

Chinese AI lab Moonshot released Kimi K3.1 Pro, a 3-trillion-parameter-class model it calls open-weight. On Terminal-Bench 4.0, a coding-agent benchmark, it scores 65.7%, ahead of OpenAI's GPT-6 Astra (58.2%) and just behind Claude Opus 5.5 (66.4%), though Moonshot ran that test inside Claude Code, Anthropic's own coding agent. One catch. Today's release runs through Moonshot's hosted apps and API only. The open weights, the part you can run yourself, are promised by November 4.

OpenAI ships GPT-6.1 Terra, a screen-driving model, into Codex at half GPT-6.1 Sol's price

OpenAI released GPT-6.1 Terra, a model built to read screens and operate apps. In Codex, its changelog says, it reasons about images and screens, then acts across your browser, desktop and apps. It costs $1 per million input tokens and $5 per million output, half GPT-6.1 Sol's list price per token, with the same 1.05-million-token context window. It's in Codex and ChatGPT Work today on paid plans, and in the API. It isn't in plain ChatGPT chat.

A critical Brotli bug threatens most Node.js and Python apps

OpenAI's Codex Security team, named for its Codex coding tool, found a critical bug in Brotli, the compression library inside Node.js, Python and most software that unpacks web traffic. Rated 9.2 of 10, it lets a malicious response corrupt memory, up to remote code execution. Node.js patched it October 21. A proof of concept followed. No attacks in the wild yet. Below Node 22.23.4, 24.22.1 or 26.12.1, or Python's Brotli under 1.2.1? Patch now.

JetBrains open-sources ThinkRail, a worktree-per-agent coding IDE

JetBrains open-sourced ThinkRail, an early-stage "incubator" project it calls an agentic IDE: every coding session runs in its own git worktree, so parallel agents never collide on your working tree. It's built around an open-source coding agent called "pi," a separate project, not JetBrains' own, which does the actual coding while ThinkRail owns the editor and workspace around it. Project intent lives in a "spec graph" beside your code that the agent reads and updates, instead of losing context between sessions. It's free, Apache-2.0 licensed, and runs on macOS, Windows and Linux.

GitHub Copilot turns Teams bug reports into pull requests

GitHub shipped a Copilot-powered app for Microsoft Teams: @mention @github on a bug report in any channel or chat, and Copilot investigates, fixes it, and opens a pull request without anyone leaving the conversation. It replaces the old "GitHub notifications" app. Microsoft retires that app November 16. Teams running it need to install the new, consolidated one to keep Copilot working inside Teams at all.

Claude's chat and Cowork merge for Enterprise plans too

Claude Enterprise plans now get the unified chat-and-Cowork experience Pro and Max subscribers got in September: Cowork, Claude's mode for coding and file tasks, now shares one conversation with chat, and keeps running in an Anthropic-managed cloud environment after you close your laptop. Admins get a new "Efficient mode" and role-based access control. Pro and Max users see no change.

Also worth your time

Ones to watch

Early, unverified: engagement, not endorsement.

Corgi: a terminal multiplexer for running several coding agents side by side, Show HN, 139 points

Know someone who'd want this in their inbox? Forward it — that's how this grows. And if we got something wrong, or you think we buried the real story today, hit reply. A person reads every one.

The New Way is written with AI. It gathers the day's stories, checks them against their sources and drafts every summary. A person decides what runs and reviews every issue before we hit send.

Anthropic files for a Nasdaq IPO and calls Claude Code its defining product

What changed this week