Skip to content
LiveNext 14:14:10
Today's five/N°44 · Thursday, November 19, 2026/Filed 03:48 ET

AI news today: Google ships Gemini 4 Neon, Anthropic prices its IPO at $83

A cheaper midtier Gemini changes your model bill, a public Anthropic changes who it answers to, and a hijacked Python package needs cleaning up today.

The five · ranked by consequence

Permalink ↗
  1. Google's Gemini 4 Neon undercuts Carbon on price

    Google released Gemini 4 Neon on Wednesday at $3 per million input tokens and $15 per million output, with a 1 million token context window. Once Carbon's price doubles on Dec. 1, Neon is the Gemini most builders should test first.

    Why it matters If you run agents or coding work on Gemini, Neon is cheaper than Carbon at list price and Google says it beats it on coding. Rerun your Carbon jobs on Neon before Carbon's introductory price ends on Nov. 30, and compare the bill.

    SRCGoogle

  2. Anthropic prices its IPO at $83, above its range

    Anthropic priced 980 million shares at $83 on Wednesday, above its $70 to $78 range, and starts trading on Nasdaq as ANT on Thursday. Demand held through a shaky stretch for tech, and Claude's maker now answers to shareholders every quarter.

    Why it matters Nothing in your Claude account changes today. What changes is that Anthropic's prices, launches and model retirements will now be read through earnings calls, so check this week that your stack can switch models without a rewrite.

    SRCAnthropic

  3. Microsoft's agent-framework on PyPI was hijacked: rotate secrets

    Wiz says versions 1.26.1 and 1.26.2 of Microsoft's agent-framework Python package carried a credential stealer for about an hour on Nov. 17, published with what appear to be stolen Microsoft signing credentials. If you installed either, assume your secrets are gone.

    Why it matters If any CI job, container or laptop pulled agent-framework on Nov. 17, pin to 1.26.0, check for the planted .pth files and rotate every secret that machine could reach today. Pinning to a trusted publisher did not protect anyone this time.

    SRCWiz

  4. California proposes power to shut down frontier AI models

    Gov. Gavin Newsom released recommendations on Wednesday that would let California order an emergency shutdown of a frontier AI model and cap a lab's recursive self-improvement. They are proposals for the legislature, but the self-improvement cap is a new kind of rule.

    Why it matters If you build on a frontier API, these rules would reach you through your provider: verified shutdown plans, 48-hour incident reports and a state able to order a model offline. Watch the bills in December, because they decide how much of this becomes law.

    SRCGovernor of California

  5. Palo Alto Networks buys Applied Compute for $4.5B

    Palo Alto Networks agreed on Wednesday to buy Applied Compute, which tunes and serves open-weight AI models, in a $4.5 billion all-stock deal. A security giant now wants to host your models, not just guard your network.

    Why it matters If you fine-tune or serve open-weight models, a security vendor just joined the clouds and inference startups competing for that work. Expect it bundled into security contracts, and ask any vendor who can see your weights and data and what happens when you leave.

    SRCPalo Alto Networks (SEC Form 8-K)

Prompt of the day

Find which jobs belong on a cheaper midtier model

› I want to cut my AI model costs without losing quality where it matters. Help me decide which of my tasks belong on a midtier model and which need a top model.

Here are the tasks I run with AI, roughly how often, and what a bad answer costs me:
[list each task, e.g. "summarize support tickets, 300 a day, low cost if wrong" or "refactor production code, 10 a week, high cost if wrong"]

Here is what I pay now, per million tokens (input / output):
[current model and price]
And the cheaper option I am considering:
[candidate model and price]

Do this:
1. Sort my tasks into three buckets: safe on the cheaper model, test before switching, keep on the top model. Give one line of reasoning per task.
2. For the "test before switching" bucket, write me a five-example test set per task: the input I should use and what a passing answer must contain.
3. Estimate my monthly savings if the first bucket moves now, showing the arithmetic. Ask me for token counts if you need them instead of guessing.
4. Name the one warning sign that should make me move a task back to the top model.

Keep it practical and short. If my list is missing information you need, ask before you answer.

Works in any AI chat. In Magai, run it against several models at once.

Every day, on the record.

Each square is an edition. Click any day to read the five that mattered then.

Editions
44
Stories ranked
220
Topics tracked
10
Terms defined
206
MayJunJulAugSepOctNov
EditionNo editionTodayFull archive →

Speak fluent AI.

Every term our stories use, defined in plain English and linked wherever it appears.

The glossary →