Google ships Gemini 4 Hydrogen for everyday coding at $1.50 per million input tokens
Hydrogen's price is 25% below Google's bigger Gemini 4 models until Dec 1, when theirs doubles. Four more stories inside.
Today leads with Google. Its new model drew two Hacker News threads past 400 points each. If you send routine work to a cheaper model, it's one more to compare on price. Also today: Claude Code can now write JavaScript to call its own tools. Anthropic opened a private beta for running Managed Agents, its agent service, in your own cloud. Claude sign-ins failed for two hours this morning. And OpenAI launched agents that draft security patches.
What changed this week
Google ships Gemini 4 Hydrogen for everyday coding at $1.50 per million input tokens
Google released Gemini 4 Hydrogen on Nov 12, the fast, cheaper member of its Gemini 4 family, built for "everyday coding and agentic workflows." It costs $1.50 per million input tokens and $7.50 per million output, in the Gemini API and, for paying subscribers, in Antigravity, Google's coding app. Google says it scores 57% on Terminal-Bench 4.1, a test of agents working in a terminal, against 18.2% for Gemini 3.8 Flash. Artificial Analysis, an independent benchmarking firm, ran Hydrogen's top setting: 54 on its index, $3.16 per task, and "very verbose." Developer Simon Willison notes Hydrogen is only 25% cheaper than its bigger siblings, Carbon and Argon, until their price doubles Dec 1. It also costs 15 times Claude Haiku 5.5's $0.10/$0.50. Try it on a real task before you switch.
Koray Kavukcuoglu
Claude Code 2.1.323 adds Code Mode: Claude writes JavaScript to call its tools
Anthropic published Claude Code 2.1.323 to npm on Nov 12. Its changelog opens with Code Mode, "an experimental opt-in in which Claude writes JavaScript to call its tools." Admins can turn it off for a whole organization. The same release drops the old v1 client for MCP, the plug standard agents use to reach tools; a setting that asked for v1 now gets v2 in a compatibility mode. The linked docs page returned a 404 on Nov 13. Nothing changes until you opt in.
Source: raw.githubusercontent.com
Anthropic lets companies host Claude Managed Agents in their own cloud, in private beta
Managed Agents is Anthropic's service for long-running Claude agents. It supplies the harness that calls Claude, the sandboxes, credentials and session history. Now companies can run "every part of the service" in their own cloud or on-premise, in private beta. Model calls can go to Anthropic, Amazon Bedrock, Google, Microsoft Foundry or "your own gateway," and session history stays in your own database. Anthropic pictures agents "fixing broken builds" unattended. Access goes through your Anthropic account team.
Claude Code and claude.ai sign-ins hit errors for two hours, Anthropic's status page says
Anthropic's status page opened an incident at 09:58 UTC on Nov 13: "an issue with users authenticating to Claude.ai and Claude Code." It lists the Claude Console and Cowork as affected too. A fix went out at 11:05, and the incident was marked resolved at 11:58, two hours after it opened. The page names no cause. Hit a sign-in error this morning? Check status.claude.com before you debug your own setup.
OpenAI launches Daybreak Defense Factory — agents that find flaws and draft fix pull requests
OpenAI launched two security products on Nov 12. Daybreak Red Team, in early access, sends agents to test a company's web apps for flaws. Daybreak Defense Factory scans codebases and drafts patches. Its agents are built on GPT-6 Cyber, OpenAI's security model, The Register reports. "This way engineers get a merge-ready pull request with the patch and a test that shows that it works," said OpenAI's Nancy Wang, who leads cyber-applied AI. Engineers still approve each fix.
Jessica Lyons
Also worth your time
- Anthropic's Claude Meetings joins Google Meet calls and builds docs while you talk, in beta on Team and Enterprise plans
- Jane Street is trialing coding interviews with less code-writing and more design and review
- Edge0 open-sources Edge0-145B, a 145-billion-parameter model it says runs on a laptop
- Conductor, the app for running teams of coding agents, told users Nov 4 it was joining Meta and winding down. It now says "We'll still be developing Conductor," without saying whether Conductor Cloud still shuts down Nov 20
- OpenAI's Codex lead Tibo Sottiaux says a banked usage-limit reset, usable whenever you like, "has been activated and should land in everyones accounts sometime soon"
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.
