Tech Brew Ride Home - New Gemini

Starting point is 00:00:00 This episode is brought to you by Accenture. When your advertising operations fall out of sync, everything else follows. Spotify and Accenture are working together to reinvent the rhythm of ad sales, using automation, analytics, and smarter workflows to simplify campaign delivery and access better data across the business. The result? Less time spent on operations, more time connecting brands with the moments and fandoms that matter most. Learn more at Accenture.com slash Spotify.

Starting point is 00:00:30 Welcome to the TechBrew Ride Home for Tuesday, July 21st, 2026. I'm Brian McCullough today. Google launched Gemini 3.6 and started pre-training Gemini 4. China weighed tightening AI export controls. Bessent said the U.S. would probe Chinese model distillation. Apple prepped a Klarna-backed leasing program and OpenAI passed 10 million codecs and work users. Here's what you miss today in the world of tech. Every day, shareholders meet to discuss important matters about the companies you invest in. Now you can make your voice heard too. Vanguard investor choice makes it easy to set your

Starting point is 00:01:13 proxy voting preference for your eligible Vanguard index funds, whether you hold a Vanguard fund directly or through another brokerage firm. All it takes is a few clicks to select your proxy voting preference and be heard on important shareholder topics like executive pay and director elections. Visit vanguard.com slash investor choice to learn more. It's your shares. It's your voice, it's easy. Vanguard investors own shares of their index funds and those funds own shares of the companies they invest in, Vanguard Marketing Corporation distributor. I told you they had been a bit too quiet. Google today launched Gemini 3.6 Flash, 3.5 Flashlight and 3.5 Flash Cyber and says it has started its most ambitious pre-training run yet set for Gemini 4. But most importantly,

Starting point is 00:02:05 for these new models, Google says Gemini 3.6 Flash delivers better coding, knowledge work, and multimodal performance while reducing output token usage by up to 17%. Even if you charge more for tokens, if you have to use less of them, it's a win, right? But also, Google has priced Gemini 3.6 Flash lower than 3.5 Flash at $1.50 per million input tokens and $7.50 per million output tokens. Gemini 3.5 Flashlight is priced at 30 cents and $2.50 respectively, quoting 9 to 5 Google. Compared to 3.5 Flash, the new model consumes 17% fewer output tokens per the artificial analysis index, while taking, quote, fewer reasoning steps and tool calls to accomplish multi-step workflows. At the same time,

Starting point is 00:02:55 it's priced lower at $1.50 per million input tokens and $750 per million output tokens versus $9 per million output tokens. In terms of coding performance, Gemini 3.6 Flash, quote, delivers higher precision with fewer unwanted code edits and reduced execution loops than its predecessor. Specifically, it generates higher quality and more reliable production-ready code, as seen in Deep Sween, 49% versus 37%, and shows significant improvement in ML Research as seen in MLE bench at 63.9% versus 49.7%. For knowledge work, the model.

Starting point is 00:03:32 The model scores 1421 on GDP Val AA versus 1349. Computer use capabilities go from 78.4% to 83%. The knowledge cutoff date for this new model finally advances from January 2025 to March 26. The company also announced Gemini 3.5 Flashlight for high throughput and low latency tasks like Agentic Search and Document Processing. Google says it offers significantly better quality than 3.1 light from March while pricing is 30 cents per million input tokens and $2.50 per 1 million output tokens. The final announcement today is Gemini 3.5 Flash cyber for finding and fixing security vulnerabilities. Flash serves as the foundation given its performance and efficiency with the goal of detecting, validating, and patching code security issues at scale and at a lower price per token than larger

Starting point is 00:04:27 models. Google's code mender tool uses multiple 3.5 Flash cyber agents. Access is initially for governments and trusted partners, quote, as part of a limited access pilot program. This will give frontline defenders a head start in finding and fixing critical vulnerabilities before they can be exploited while mitigating against broader misuse. Gemini 3.6 Flash and 3.5 Flashlight are available today in the Gemini app. With the latter model also coming to search, developers can access them through Google anti-gravity, AI studio, and Android Studio. Looking ahead, Google reiterates that. that Gemini 3.5 Pro is currently testing with partners with broad availability as soon as it's ready.

Starting point is 00:05:07 Meanwhile, the DeepMind team is, quote, already focusing on building the next generation of models. We have already started our most ambitious pre-training run yet for Gemini 4 and can't wait to share more, end quote. Now, quoting CNBC, Gemini 3.5 Flash Cyber is designed to detect and patch software vulnerabilities and will initially be available only to governments and trusted partners through a limited access pilot. but Google said the specialized model runs at a lower price per token than other larger models. That could help Google narrow its cybersecurity gap with Anthropic, which has built an early lead in automated code defense. The broader lineup reflects Google's bet that price and efficiency can help offset its slower timing

Starting point is 00:05:48 in several key product categories. Artificial analysis data shows Gemini Flash already undercuts comparable models from Anthropic, OpenAI, and Chinese rivals on cost. according to the company, Gemini 3.6 Flash, the stronger of the two new models is cheaper per task than GPD 5.6 Terra Max, Kimi K3 and Quinn 3.7 Max, while 3.5 Flashlight costs just a fraction of that. The rollout comes on the eve of alphabet earnings and as Chinese rivals gain momentum. Moonshot AIs Kimi K3 grew enough demand that the company limited new subscriptions and API access because of capacity constraints, while Alibaba is only teasing Quinn 3.8 Max, which it said

Starting point is 00:06:31 trails only Anthropics Fable 5 in overall performance. That demand highlights the other side of the AI race. Building a competitive model is only part of the challenge. Companies also need enough computing capacity to serve at scale. Google has a potential advantage through its custom chips, cloud infrastructure, and ability to design models and hardware together, although the company has face capacity constraints of its own, end quote. Meanwhile, there continue to be signs that no sooner than open-weight Chinese models have had their moment again, then the window may shut on them again. Quoting the F.T.

Starting point is 00:07:15 Chinese regulators are considering tightening export controls on artificial intelligence and semiconductor technologies as the U.S.-China rivalry intensifies in cutting-edge AI. Regulators led by the Ministry of Commerce have been consulting leading domestic AI and chipmaking groups on how to prevent China's advanced technologies and start-ups from being acquired by the West, according to two people involved in the discussions. MFCOM talked to AI companies including Alibaba, ByteDance, and Jipu, on limiting the transfer of key data for the training of their models overseas, as well as allowing their model weights to be downloaded by foreign users, the people said.

Starting point is 00:07:52 China would still let overseas customers access the models and services, however. China's AI Lab Moonshot last week released its Kimmy K3 model that exceeded Anthropics flagship Opus 4.8 by most of the benchmarks, demonstrating that China has significantly narrowed its gap with the U.S. in terms of frontier AI. The leading Chinese models from Moonshot as well as Deepseek are so-called Openweight, which allow users to download to their local servers and customize to their particular needs. Meanwhile, leading U.S. models from Anthropic and Open AI are closed. MFCOM has also sought views on possible restrictions that would prevent overseas chipmakers including Qualcomm and TSMC from producing advanced semiconductors based on designs developed by

Starting point is 00:08:36 Chinese companies such as Huawei, Alibaba, and ByteDance, according to the people. Potential restrictions could also be imposed on the overseas acquisition of strategic technology groups in areas such as Agentic AI, the people said. This is mainly to address a loophole that Beijing believes led to META's $2 billion acquisition of Manus, a deal that was subsequently ordered to be unwound by Chinese authorities. The new measures could be incorporated into the next revision of China's catalog of technologies prohibited or restricted from export, the people said, reflecting Beijing's growing confidence that it has established a global lead in some areas of AI. Most of the proposals are still

Starting point is 00:09:16 under discussion with regulators weighing industry feedback before making a final decision, according to the people. They added that tech companies have told regulators that some of these tighter measures would slow down their AI development and hurt China's potential to win the technology race. The catalog is one of China's three main export control regimes, alongside two control lists covering dual-use items. Its most recent revision in 2025 added several lithium-ion battery manufacturing technologies to the list of restricted exports, building on existing controls over strategically important technologies such as rare earth extraction and processing. The update being discussed and negotiated would be the most significant in years.

Starting point is 00:09:54 according to the people, end quote. Now, if the window doesn't shut on one side of the Pacific, it could shut on the other side just as easily. The American frontier models don't like being undercut on price, especially when performance is increasingly comparable. And they have been claiming for a while now that the only reason the Chinese models can leapfrog them is because of distillation.

Starting point is 00:10:20 Basically, allegedly using the U.S. frontier models to reverse engineer in a way and train competitive Chinese models. Quoting CNBC, U.S. Treasury Secretary Scott Bessent on Tuesday said the Trump administration will look into whether Chinese artificial intelligence models have been distilled from American models stating that the government does not support IP theft. If we see, especially that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft. Bessent told Fox Business's mornings with Maria on Tuesday. Bessent said the technical term for this theft is called distillation, which is an AI training

Starting point is 00:10:59 method where a small, less capable model is built using outputs from an existing stronger model. Anthropics sent a letter to the U.S. Senate Committee on Banking Housing and Urban Affairs last month, alleging that the Chinese tech company, Alibaba, had carried out, quote, the largest known distillation attack against it to date. The U.S. and China are planning to hold talks around AI in September, according to a report from Reuters on Tuesday, Bessent will represent the U.S. during the discussions, the report said. We are finding watermarks of our U.S. large language models on many of the Chinese models, and that's unacceptable, Bessent said on Tuesday.

Starting point is 00:11:33 We are going to be looking at that in the coming days or weeks, end quote. The Hulu original series Furious is coming to Disney Plus, Starring Emmy Rossum, Furious follows FBI agent Alice Black on the hunt for a mysterious and calculating serial killer. Both walk their own paths toward justice, and as their lives start to intertwine, the line between right and wrong begins to blur. Don't miss the three-episode premiere of the Hulu original series Furious on July 27th, only on Hulu on Disney Plus. When critical company knowledge isn't documented, there's a major reason. ripple effect. Work becomes inconsistent. Tools don't get adopted and knowledge walks out the door when someone leaves. Thankfully, our sponsor, Scribe, was built to fix that. Their workflow

Starting point is 00:12:29 AI platform is trusted by nearly half of the Fortune 500 to capture workflows in real time. Here's how it works. You turn on the Scribe browser extension or desktop app, do a process as you normally would, and Scribe will build a guide as you go. It automatically redacts sensitive information and even suggest improvements to your existing workflows. To see what Scribe could look like for your org, head to Scribe.how slash ride home and mention ride home for your first month of Scribe capture free on select plans. That's S-C-R-I-B-E dot how slash ride home. I thought they already had something like this, but Mark German says Apple plans to launch Apple upgrade a leasing program for iPhone, iPad, Mac, and Apple Watch in partnership with Klarna in the U.S. next week.

Starting point is 00:13:16 Quoting Bloomberg, the new service plan to launch July 28th will support most iPhone, Mac, iPad, and Apple Watch models. According to people with knowledge of the matter, Apple is partnering with Klarna as the financial backer for the program, said the people who asked not to be identified discussing unannounced initiatives. The initiative will launch initially at least in the U.S. and will be available both in its physical retail stores and online. Shares of Klarna jumped as much as 11% on the news. The Apple upgrade service will work like a subscription, and users can pay off the device early during their term, upgrade early to a new model, or keep it at the end of the leasing period. In certain cases, those transactions will incur an additional fee. Like a car lease, it can also be returned at the end of the term.

Starting point is 00:14:03 Leases for iPhones and Apple Watches will run for 24 months, while Mac and iPad plans will run for 36 months. Users will need to be approved for the program via a soft credit check. The new program will launch weeks after Apple raised prices by hundreds of dollars across Macs, iPads, and other devices in response to an industry-wide memory shortage. Apple is also widely expected to hike iPhone prices when the new models are unveiled in September. Apple plans to advertise the program as a way to have lower payments versus current financing programs. The company is planning to discontinue new enrollments for its current iPhone payment programs, the iPhone upgrade program, and standard financing to make way for the Apple upgrade initiative. Unlike the iPhone upgrade program of today, Apple upgrade doesn't include Apple Care. Certain devices, including the Apple Watch SE, entry-level iPad, iPhone 16, and MacNeo aren't eligible.

Starting point is 00:14:55 Business and education purchases are also excluded. Apple had been eyeing its own in-house iPhone and hardware subscription program before canceling the plan in 2024 by partnering with Klarna. Instead, Apple gets the benefit of offering such a way to buy its products without taking on the financial responsibility, end quote. Back to AI real quick. OpenAI says it now has 10 million people using Codex and ChatGTPT work nearly doubling usage from earlier this month when the company announced ChatGPT work, quoting Implicator AI. ChatGPT work is the part of ChatGPT built for longer research, analysis, and finished

Starting point is 00:15:41 deliverables, while Codex remains dedicated to software development according to OpenAI's help center page. The count arrives less than two weeks after the implicators July 10th coverage noted OpenAI's earlier claim that Codex alone had more than 5 million weekly users. OpenAI says chat GPT itself has more than 900 million weekly users, a much larger base than the agent products now being measured. OpenAI did not separate Codex users from chat GPT work users in the combined tally. They also do not say whether the count means weekly active users, monthly active users, people who started one agent task or people who completed reviewed work. Open AIs help page says work and codex draw from the same agentic usage and credit pool,

Starting point is 00:16:26 and that the amount used depends on task size, complexity, model, and surface. The June data covered codex only in a June 25th paper researchers using Open AI data reported that weekly active codex users increased more than fivefold between January 1st and June 1st. In the last 28 days of the observation period, 17.3% of organizational users who were active on chat GPT or codex used codex compared with fewer than 1% of active individual users. The study also warned that opening eye workers were an unusually favorable environment with high model familiarity, low marginal cost, and broad internal support, end quote. Finally, today, Garmin has launched the Circa Smart Smart.

Starting point is 00:17:15 band, a $200 screenless fitness tracker that monitors more than 80 activities with an up to 10-day battery life to compete with offerings from Whoop, quoting the verge. Garmin announced a new smart band today designed to track advanced fitness and wellness features without a screen to eliminate distractions and push its battery life up to 10 days between charges. Similar to competing products like the Whoop 5.0 and Google's recently announced Fitbit Air, the Circa uses a combination of heart rate, blood oxygen, skin temperature, and motion sensors to monitor your health and activities around the clock, but without any subscription fees. The Garmin Circa is available now through the company's online store for $190.99 and comes with an adjustable fabric strap in several different colors that can be worn on the wrist or as an armband. Google's Fitbit Air is $99 with an additional $99 per year annual Google Health Premium subscription that unlocks advanced feature.

Starting point is 00:18:11 while Woop's latest hardware is only available with subscriptions that start at $199 per year. The smart band can track more than 80 different activities, such as running or yoga that are selectable through the accompanying Garmin Connect app or using its single-side button as a shortcut to your favorite. Through the app, you can monitor your training progress, including the effectiveness of a workout and whether your body is ready for your next training session or would benefit more from a day of rest. When worn day and night, the Circa can keep tabs on health metrics, including your heart rate and blood oxygen saturation that can help detect conditions such as asthma and sleep apnea.

Starting point is 00:18:49 In the morning, it can provide a detailed breakdown of your sleep stages, breathing, and heart rate that contribute to an overall sleep score so you can compare the quality of your rest to previous nights. Women can also use the Circa's skin temperature monitoring to track their menstrual cycles for, quote, better period predictions and past ovulation estimates, end quote. If you check my Twitter at Brian MCC, you will see that our podcast puppy Archie has a paw problem and is wearing a cone of shame. Talk to you tomorrow.

Tech Brew Ride Home - New Gemini

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.