@10xmylife: Ultimately, the root cause of weak model capability is simply a compute problem
Summary
The author believes that the essence of insufficient model capability is a compute issue, Gemini 4 helped Google make a comeback, and the gap between domestic and top overseas models ultimately hinges on whether the compute bottleneck can be solved.
View Cached Full Text
Cached at: 11/16/26, 06:24 PM
At the end of the day, the root cause of insufficient model capability is really just a compute problem.
Gemini — the “American Doubao” that was mocked for a year or two — has made a complete comeback with Gemini 4. Even though Google strictly restricts reverse proxying so it can only be used inside that 💩-tier antigravity, I’m still forcing myself to use it, because the model is genuinely great.
What does this tell us? There’s no mystical algorithm moat, and no absolutely insurmountable bottleneck. In the end, it all comes down to two hard metrics: compute and time.
Is there a big gap between domestic models and top overseas models right now? Yes, a significant one.
But what about the future? It’s really hard to say. As long as the compute bottleneck problem can be solved, domestic teams are equally capable of delivering top-tier models on par with OPS 5.5 level.
Unfortunately, compute happens to be the hardest problem to crack right now.
Similar Articles
@10xmylife: Yeah, I’m Dario, and I’m anxious too, because large models have no moat at all. It’s purely compute + time. Google shaking off its “America’s Doubao” label with Gemini 4 proves it. With enough compute and time, you can build a very good model just the same; there are no technical bottlenecks to solve. So what should A\ and OpenAI do? Monopolize compu…
A tongue-in-cheek commentary argues that large language models have no real moat and are built simply by piling on compute and time, with Google proving the point through Gemini 4. It jokingly suggests that Anthropic and OpenAI will respond by monopolizing compute and lobbying regulators to restrict NVIDIA.
@emollick: The challenge for Google post-Gemini 4 will be what they do with a good model. Anthropic & OpenAI show we are moving to…
Ethan Mollick 分析认为,Gemini 4 之后 Google 面临的挑战是如何利用好一个优秀模型,因为 Anthropic 与 OpenAI 正展示出向单一界面加编排式代理的方向演进,而 Gemini 3 时代碎片化的产品策略将难以为继。
@sitinme: A 26M parameter model can do Function Call, and is even stronger than Qwen-0.6B? This team's out-of-the-box approach is too wild! Nowadays, large models have ever-growing parameter counts, but one question has never been seriously considered: does calling a tool really need hundreds of billions of parameters? Think about it, when you say 'Check today's...'
The Cactus team distilled Gemini 3.1 into a specialized model called Needle with only 26M parameters, specifically for Function Call. Its performance surpasses Qwen-0.6B, demonstrating the potential of small models in tool calling scenarios.
Gemini 4 Crushes Benchmarks, But Google Employees State The Model Struggles With Real Work
Google's Gemini 4 posts strong benchmark results, but internal employees report the model struggles with real-world coding tasks and practical work, raising concerns it may lag behind Anthropic and OpenAI's next-gen releases.
@RookieRicardoR: Domestic models break through again, matching top models like Claude 4.6 and Gemini 3.1 Pro. Just tested Qwen3.7-Max, sharing some real thoughts. Last night I topped up as soon as the API went live and chose three tasks (see video) to test Qwen3.7-Max's frontend capabilities…
The user tested Qwen3.7-Max and believes it matches top models like Claude 4.6 and Gemini 3.1 Pro in frontend, computing power, and Agent capabilities. Its reasoning ability has significantly improved, and with monthly iteration speed, it has become a first-tier domestic model.