Google and Tencent made nearly identical moves with their models: focusing on small models, efficiency, inference speed, low cost, and ecosystem integration—prioritizing intelligent density over peak intelligence. Google launched three models, with Gemini 3.6 Flash as the star. Tencent officially released HunYuan Hy3 on July 6. Neither pursued the most expensive flagship models. Google’s 3.5 Pro remains unreleased; Bloomberg reports it failed internal benchmark tests. Tencent’s Hy3 has 295B total parameters but activates only 21B via MoE, consuming just 7% of compute per inference. This is fascinating. Gemini 3.6 Flash’s output price dropped from $9 to $7.5, while speed surged from 165 to 304 tokens/s—nearly doubling. The intelligence score remains at 50, but it uses fewer tokens; same benchmark costs fell from $1,041 to $727—a 30% reduction. Hy3 is even more aggressive: its office agent task success rate jumped from 72% to 90%, saving 47.4% of tokens on documents and 49.0% on PPTs. Now, neither company is trying to prove which model is smarter. They’re competing for something far more valuable: the default entry point. Intelligence only needs to be sufficient—the real value lies in embedding the model into products users already rely on. Google has Copilot, Workspace, and Antigravity. Tencent has WeChat, Yuanbao, WorkBuddy, Tencent Docs, and gaming platforms. When you run workflows in WorkBuddy or generate PPTs with a single prompt in Yuanbao, that experience vanishes outside Tencent’s ecosystem. Google embedded Flash into Copilot, making it the default tool for global developers writing code. For platform holders with distribution channels, the optimal strategy isn’t topping leaderboards—it’s making their model the user’s unconscious default choice. Intelligence is becoming a commodity; integration is the real barrier. This is tough for independent model vendors: competition is homogenized, switching costs are low. Higher scores don’t matter—advanced users are limited, and average users won’t switch because switching makes little difference in daily tasks. Unless you achieve clear dominance in a specific vertical capability, breaking through platform lock-in is nearly impossible. Over the next year, victory won’t hinge on model parameters—but on agent stability. Whoever delivers more reliable, efficient, and error-free multi-step workflow pipelines will win. Google and Tencent, with their control over traffic entry points, are on the right path—and their monetization will proceed more smoothly.