THE FRONTIER BRIEF


  Google DeepMind says Gemini Omni 1.2 Flash can create and edit videos up to 15 seconds long, up from 10 seconds with Gemini Omni 1.1.͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ ͏‌ 

  
  
    
      

        
        

          
          
            
              
                
                  
                
              
              
                THE FRONTIER BRIEF
              
              
                2026-11-06 · Signal, Not Noise
              
            
          

          
          
             
          

          
          
            
              
                ■ The Big Story
              
              
                Google cuts the price of making short AI videos
              
              
                Google DeepMind says Gemini Omni 1.2 Flash can create and edit videos up to 15 seconds long, up from 10 seconds with Gemini Omni 1.1. It is generally available through the Gemini API and Gemini for business. The company says 720p video costs $0.07 per second, 30% less than Gemini Omni 1.1. Marketing teams now have a clearer cost for testing short video variations.
Google DeepMind says teams can choose clip lengths in one-second steps from three to 15 seconds. That makes a difference when an ad needs a shorter opening or a product explainer needs several cuts: teams can pay for the length they need rather than defaulting to the longest clip. Separately, Anthropic says Personal Finance in Claude is rolling out in the US, letting users ask about connected bank and card accounts through read-only access.
              
              
                
                  Why It Matters for Business Leaders
                
                
                  If short video is a recurring expense, compare your agency’s draft-production charges with the new per-second cost before renewing that part of the contract.
                
                
              
              Sources: [Google DeepMind] [Anthropic]
            
          

          

          
          
            
              
                ■ Frontier Watch
              
              
                
        
          
            OpenAI restricts access to its stronger security model [The New Stack]
            
              The New Stack reports that OpenAI launched GPT-6 Cyber and rates its ability to find software weaknesses and develop ways to exploit them as “critical,” up from “high” for its predecessor. OpenAI says access is limited to selected approved Daybreak Red customers.
            
            In Plain English: A more capable digital lock-picker is being issued to a restricted group, not put on an open shelf.
            
              Why it matters: Security teams cannot budget around immediate general access; ask existing testing vendors how they will handle the stronger tools becoming available to approved users.
            
          
        
        
          
            Mistral gives companies more places to run AI workers [Mistral AI]
            
              Mistral says it is expanding Studio with managed agents and apps. It says teams can deploy them in the cloud, on their own private infrastructure, or across both.
            
            In Plain English: A company can choose where its AI workers operate, much as it chooses whether to keep files at its own office or with a storage provider.
            
              Why it matters: That deployment choice gives teams with strict data-location requirements another option when comparing AI vendors.
            
          
        
        
          
            OpenAI’s president raises the prospect of a public listing [TechCrunch]
            
              TechCrunch reports that OpenAI president Greg Brockman said the company could be ready for an initial public offering in 2027. TechCrunch says he also pointed to the company’s gains in the enterprise market.
            
            In Plain English: OpenAI is discussing a possible step toward selling shares to the public, while competing for more business customers.
            
              Why it matters: A possible listing is not a reason to accept a longer software commitment today; price and exit terms still determine the value of a contract.
            
          
        
              
            
          

          

          
          
            
              
                ■ The Street Says
              
              
                The sharp selloff has put the cost of building AI infrastructure under scrutiny, even as AI companies pitch larger roles for their products inside businesses.
              
              
                
        
          
            
              ↳ Re: AI infrastructure spending and falling chip stocks
              skeptical/mixed
              
                Peter Boockvar, quoted by CNBC →
              
              
                “if there is a slowdown in the actual development and the buildings and the infrastructure and so forth”. Business buyers should separate an AI vendor’s current service from promises that depend on future construction.
              
            
          
        
        
          
            
              ↳ Re: data-center construction after the elections
              frustrated/bearish
              
                Dina Powell McCormick, quoted by CNBC →
              
              
                “building data centers is ‘going to be harder’ following the elections”. Companies negotiating long-term computing commitments should ask vendors what happens if planned capacity arrives late.
              
            
          
        
        
          
            
              ↳ Re: market stabilization and the Anthropic IPO
              cautiously optimistic
              
                Max Kettner, quoted by Bloomberg →
              
              
                “Let's see in the next two weeks if that goes through, that will give us some sort of relief”. Owners should not let a hoped-for market rebound set the timing of an AI purchase.
              
            
          
        
        
          
            
              ↳ Re: Perplexity Teammates taking multiple roles at work
              positive
              
                Aravind Srinivas, quoted by The Verge →
              
              
                “Over time, we envision every org having hundreds of teammates working for each person”. Treat that as a vendor’s ambition, not a staffing assumption for next quarter.
              
            
          
        
              
              One more thing
Meta’s Muse can now take calls as an animated avatar, ahead of a Tamagotchi-like wearable. The avatar era has arrived with a pocket-pet side quest. TechCrunch →
            
          

          

          
          
            
              
                ■ Repo of the Day
              
              
                
                  
                    NVIDIA / auto-ontology
                  
                
                
                  ⭐ 8,048 stars · +2,679 today
                
                
                  
                    Why It Matters
                  
                  
                    It offers a way to make answers drawn from company data easier to trace back to the information used.
                  
                
                What it does: This Python framework builds a map of relationships in data. It generates repeatable queries and retrieves relevant background for AI assistants.
Problem it solves: An assistant can miss a relevant record or return an answer that is difficult to explain when company information is scattered.
Who should use it: Data and IT teams building internal question-answering tools.
Expected outcomes: Teams can produce more consistent searches and give reviewers a clearer path from an answer back to its supporting records.
              
            
          

          

          
          
            
              
                ■ Two Steps Ahead
              
              
                Over the next 30–90 days, marketing teams using Gemini Omni 1.2 Flash can ask agencies to separate concept development and final finishing from charges for generating draft clips. The price change makes draft production easier to compare; it does not erase the value of editing, approvals or distribution.
For security procurement, GPT-6 Cyber’s restricted release makes access terms more important than a headline capability claim. Ask testing providers whether approved Daybreak Red access is included in your contract. Meanwhile, Mistral’s Studio gives businesses with data-location requirements a concrete deployment alternative to put beside cloud-only proposals.
If an internal data assistant is being scoped this quarter, NVIDIA / auto-ontology points to a different purchasing question: how will staff trace an answer to company records? Make that a requirement before comparing assistants on how fluent their responses sound.
              
              Tools & Platforms to Watch
Gemini API: Price and produce short video drafts for campaign variations.
Studio: Compare cloud and privately hosted deployment for company AI workflows.
              Further Watching & Reading
[Read · Google AI Blog] Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips
[Read · TechCrunch] OpenAI released a new cybersecurity model. Is that enough?
            
          

          
          
            
              
                
                  View in Browser
                
                 · 
                
                  All Editions
                
              
              
                THE FRONTIER BRIEF
              
              
                A HUMN Solutions Intelligence Brief · 2026-11-06

                Signal, not noise. Curated and synthesized by AI, vetted for substance.