Google releases Gemini 4 Carbon to all developers, a day after Claude Fable 5.2
An outside test ranks it fourth to Fable's first, at under a third of the cost per task. Five more stories inside.
Today leads with Google because developers now have two new top models to choose between. If you pick models for a team, the comparison starts this week. An outside test ranks one higher and the other far cheaper. Also today: Anthropic gave every paid Claude plan one free limit reset. A security firm found 35,578 Grok-built apps open to anyone. An outside count found 217 fixes among Anthropic's 37,966 bug findings. Claude got spreadsheets and whiteboards. And browser makers shipped guides that get coding agents off old JavaScript.
• Google releases Gemini 4 Carbon a day after Fable 5.2
• Anthropic gives paid Claude plans one free full reset
• Grok Build left 35,578 apps open to anyone
• VulnCheck finds 217 fixes in Anthropic's 37,966 findings
• Anthropic adds Sheets and Whiteboards to paid Claude plans
• Browser makers ship guides that modernize coding agents' web code
What changed this week
Google releases Gemini 4 Carbon to all developers, a day after Claude Fable 5.2
Google released Gemini 4 Carbon, its strongest model, to every paying developer on Nov 10, a day after Anthropic shipped Claude Fable 5.2, its own top model. Google's post doesn't mention Anthropic. Carbon had gone only to "cybersecurity defenders" two weeks earlier. Artificial Analysis, which runs models itself, ranks Fable 5.2 first of 74 and Carbon fourth on Nov 11. Carbon costs $7.76 per task there. Fable costs $26.54. Carbon's API price is $2 per million input tokens and $10 per million output until Nov 30, then $4 and $20. It's also in Cursor and GitHub Copilot. Google AI Ultra subscribers get it in Antigravity, Google's coding app, which previews a /swarm command that splits one prompt across a team of agents. Google says Carbon ranks first on CursorBench 4.0, Cursor's coding test. On the outside ranking it's cheaper, not best. Try it before the price doubles Dec 1.
Gemini 4 Carbon is rolling out today. It’s built for sustained, long-horizon tasks, reasoning deeply through complex software engineering, cybersecurity, and scientific discovery.
— Sundar Pichai (@sundarpichai) November 10, 2026
Coming globally via the Gemini API in @GoogleAIStudio and Gemini for business, and for Ultra…
Anthropic gives every paid Claude plan one free full limit reset, usable until Dec 10
Anthropic's Claude Devs account said on Nov 10 that every Pro, Max, Ultra and Team user has "one free full reset, across all models and limits," usable "anytime before December 10 at 2:00PM PT." Its reason: it fixed a few small issues that used more limits than intended, and the Fable resets did little for people already at their weekly limits. Find it in Settings > Usage, not in Claude Code's terminal. It can't be undone. Save it for the week you hit the wall.
Grok Build left 35,578 apps open to anyone with the link, Claroc finds
Security firm Claroc tried 43,551 addresses of apps made in Grok Build, where people describe an app to Grok and it writes one. Of 37,571 still online, 35,578 "opened for anyone, with no sign-in." In Claroc's test account, Publish proposed "Anyone with the link." Open apps held a sales onboarding plan from SpaceXSI, the merged SpaceX and xAI, and, per Ars Technica, 3,784 worker medical exams. SpaceXSI pulled its 24 staff apps. Claroc says the rest stayed open. Check your Grok apps' sharing.
Asaf Aprozper, Claroc Research
VulnCheck counts 217 fixes among the 37,966 findings on Anthropic's vulnerability ledger
Patrick Garrity of VulnCheck, a vulnerability-tracking firm, has counted bugs credited to Anthropic since April 17, shortly after its Glasswing bug-hunting program launched. He finds 442 CVEs, the public IDs for known flaws, in seven months, with "only 2 of the 442" used in real attacks. Anthropic's own ledger, as of Nov 2, shows "37,966 findings and just 217 fixes," and Garrity counts "7.5x more duplicates than fixes" in it. A finding isn't a fix. Treat Claude's security scans as leads to verify.
Anthropic adds Sheets and Whiteboards to paid Claude plans, built on MCP Apps
Anthropic launched Claude Whiteboards and Claude Sheets in beta on Nov 10 and made Claude Dashboards, its data-question tool, generally available as Claude Data, on "all paid plans." A second Anthropic post lets tools like Notion open beside the chat while Claude edits them. Those run on MCP Apps, an open extension to MCP, the plug standard AI agents use. Any MCP server can now bring its own interface into Claude. Build one if you ship a connector.
Browser makers ship a free guide set that steers coding agents off legacy JavaScript
Modern Web Guidance is a set of skills, instruction files a coding agent loads on demand, "stewarded by the Google Chrome, Microsoft Edge, Mozilla Firefox, and MDN teams." Its makers say agents "often default to older patterns because LLM training data contains vast amounts of legacy code." Version 1.0.0 went up on npm on Nov 10. One npx command installs it. A legacy scan flags old patterns in an existing codebase. Run it before your agent's next front-end task.
Also worth your time
- Git 2.56.1 fixes six security bugs, including code execution when recursively cloning a malicious repository
- Microsoft researchers backdoored an AI agent by reprogramming fewer than 10 dormant neurons, with no training
- Val Town, the hosted-scripts platform, stops signups and goes read-only Dec 10
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.
