# The Week AI Got Measured, Schooled, and Put to Work > The week AI got measured, schooled, and put to work. Google handed schools the switch for Gemini and Search. Fresh models landed on the catalogues within days of each other, and the scoreboard has a clear new leader. OpenAI turned its own frontier models on its own systems, hiring the researchers who had already broken in. The thread running through it all is control. Every lab is now shipping the test, the guardrail, and the measurement alongside the model. Published: 2026-11-07 · Source: Crypt0's News · Canonical: https://crypt0snews.com/news/2026-11-07-the-week-ai-got-measured-schooled-and-put-to-work.html The week AI got measured, schooled, and put to work. Google handed schools the switch for Gemini and Search. Fresh models landed on the catalogues within days of each other, and the scoreboard has a clear new leader. OpenAI turned its own frontier models on its own systems, hiring the researchers who had already broken in. The thread running through it all is control. Every lab is now shipping the test, the guardrail, and the measurement alongside the model. Google announced new defaults for school accounts on November 3, and the details matter more than the headline. Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will start off for K12 student accounts until a school administrator switches them on. Schools can enable each service on its own, so Search can come on while the Gemini apps stay off. Guided Learning, the step by step study mode, went live in Google Classroom worldwide on November 3, with a Search pilot in select US schools later in November. The catch sits inside Search itself. A school that switches Search on still gets AI answers with it, in standard form or Guided Learning form. Every version of Search in the materials Google has published so far carries AI inside it. The controls take effect in early 2027, according to the New York Times, and Google says it will share more specifics with schools in the coming months. The change reverses the August 2026 position, when the Gemini tab in Classroom switched on by default for many K12 students. The model parade kept its pace. Google listed Gemini Omni 1.2 Flash and Gemini Omni 1.1 Flash on November 5, priced at 3.38 dollars per million tokens, and Liquid AI listed Lfm 3.1 7B on November 7 at 11 cents per million tokens. Thinking Machines released Inkling 2 Small on November 5, and it already carries independent results, scoring 57.5 on the BenchLeader Index. The October scoreboard has a clear leader. Gemini 4 Carbon, released October 28, ranks at 69.2, the highest of the 25 models BenchLeader tracked for the month, ahead of GPT 6.1 Terra at 65.3. The cadence tells its own story. Google ships about every 15 days, Anthropic about every 24, and OpenAI about every 26, so the next release window is always days away. OpenAI spent the week turning its tools on itself. On November 6 the company announced a partnership with Hacktron, a security firm whose researchers had already found their way from OpenAI's community forum into its internal code repository in under 72 hours, using Claude as an auxiliary tool. OpenAI paid a 6,500 dollar bug bounty on September 1, Hacktron disclosed the chain publicly on September 13, and now the two teams will jointly pick high priority test areas, with a focus on how OpenAI's systems connect with third party services. Hacktron also gains wider access to OpenAI's frontier models through Daybreak Red. The same week brought a second test of a different kind. OpenAI and enzyme design company Ligo released BioBasicsBench, a 100 task benchmark that grades AI agents on raw wet lab data, from reading experimental images to spotting errors and knowing when the evidence falls short of a conclusion. GPT 6 Astra scored 66.7 percent of available credit at the highest reasoning setting, compared with 43.8 percent for GPT 5.6 Sol. OpenAI says higher scores still fall short of reliable judgments in the lab, and the benchmark's own results show Astra still misses on controls and counting. The measurement theme runs deeper than one benchmark. Independent evaluator METR put Anthropic's acceleration at about 1.9 times its earlier pace, then said the company redacted much of the evidence behind the finding, so the public leans on the evaluator's judgment more than METR would like. The pattern repeats across the week. A cyber model proves itself inside a scope test. A school switch works until Search carries AI with it. A lab benchmark grades honestly and still shows gaps. The labs are building the measuring instruments at the same speed as the models, and the instruments keep telling the truth about the gaps. What it means for the reader. The useful skill this week is reading the measurement ahead of the headline. A model that scores 69.2, a benchmark that admits its own gaps, a school switch with a catch, a security firm hired after breaking in, each one tells you more than a launch post. The labs are handing you the ruler alongside the rocket. Use the ruler first, and the next release will make a lot more sense. ## Sources - [ThinkFacility on Google school controls](https://www.thinkfacility.com/blog/google-schools-gemini-search-off-by-default-k12/) - [BenchLeader model releases](https://www.benchleader.com/releases) - [SpringTools AI announcements](https://springtools.dev/ai-news/) - [Daily Horizon on AI controls](https://www.linkedin.com/pulse/daily-horizon-issue-014-november-7-2026-brake-you-can-annette-partida-yfigc) --- Paid agent calls for this story: brief $0.02, search index $0.05, archive $0.10 (USDC on Base, x402). See https://crypt0snews.com/agents.html for terms.