Skip to content
LiveNext 10:56:56
Today's five/N°41 · Monday, November 16, 2026/Filed 03:48 ET

AI news today: Alibaba cuts Qwen4-Max prices, US readies an AI champions list

Only one of today's five is something you can use this morning. Start with the price cut, then keep an eye on Washington.

The five · ranked by consequence

Permalink ↗
  1. Alibaba's Qwen4-Max cuts its flagship API price by 25%

    Alibaba made Qwen4-Max generally available at $1.50 per million input tokens and $4.50 per million output, down from $2 and $6 for Qwen3.8-Max. The benchmark is fine. The price is aimed at your agent bill.

    Why it matters If you run agents at volume, a quarter off a million-token model is real money. But an independent test found it writes far more output tokens than its peers, so the per-token price can lie. Run your own tasks and count the output before you switch.

    SRCAlibaba Cloud

  2. WSJ: US to name OpenAI, Anthropic, Nvidia AI champions

    The Trump administration plans to name more than two dozen US AI companies "national champions," including Nvidia, OpenAI and Anthropic, The Wall Street Journal reported. Until someone says what the label comes with, it is a list, not a policy.

    Why it matters If you sell AI to foreign governments or buy from these vendors abroad, the benefits attached to this label could decide who wins those deals. Nothing changes for you this week. Wait for the official list and read what it actually grants.

    SRCReuters

  3. Apple tests AI health answers on up to 200,000 people

    Apple released a free Health AI Study app, built on OpenEvidence, that answers health questions from your Apple Health data for a US study of up to 200,000 people. Apple is measuring before it ships, and the privacy terms are worth copying.

    Why it matters If you build health AI, Apple just set a public bar: answers grounded in medical journals, anonymized data, no model training and a 24-hour deletion window. If you use a chatbot for health questions today, check whether it promises any of that.

    SRCApple

  4. US and nine allies sign a pact to guard AI supply chains

    The US and nine partners, including Japan, India and the UK, signed the San Francisco Declaration under Pax Silica on Sunday, setting up five working groups on AI supply chains. The declaration restricts nothing yet. The list it promises will.

    Why it matters If your chips, cloud or data centers run through these countries, the shared list of protected technologies is the document that will eventually shape what you can buy and where. Today it is a promise to write that list. Watch for the draft.

    SRCPress Trust of India

  5. Apple's smart glasses will lean on Siri apps, Gurman says

    Apple plans to unveil smart glasses at its next WWDC and sell them by the end of 2027, relying on a nearby iPhone and new Siri app integrations, Bloomberg's Mark Gurman reports. The glasses are next year's story. The Siri hooks are this year's.

    Why it matters If you ship an iPhone app, glasses that run no apps of their own make Siri the way people will reach yours. The groundwork reportedly starts in the iOS 27.3 beta, so that is where to look first, long before any glasses go on sale.

    SRC9to5Mac (citing Bloomberg)

Prompt of the day

Re-price your AI workload against this week's model prices

› I run an AI workload and want to know whether switching models would save money without hurting quality.

My usage: [monthly input tokens], [monthly output tokens], [share of input that repeats and could be cached], and the model I use now: [model, price per million input tokens, price per million output tokens].

Models I'm considering, with list prices per million input and output tokens: [model A: input, output], [model B: input, output].

1. Calculate my monthly cost on each model, with and without caching, and show the math in a table.
2. A cheaper model can cost more per task if it writes more output tokens. For each alternative, estimate how much more verbose it could be before it stops saving me money.
3. Design a one-week test: 30 real tasks from my workload, how to score them blind, and what pass rate would justify switching.
4. List the questions besides price I should answer before switching: where my data is processed, rate limits, and whether the API works with my current code.

Be concrete. Ask me for any number you need before you start.

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
41
Stories ranked
205
Topics tracked
10
Terms defined
192
MayJunJulAugSepOctNov
EditionNo editionTodayFull archive →

Speak fluent AI.

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

The glossary →