Open Models are the United Front of AI
ChinaTalk
Opinion — commentary, not a factual news event.
Chinese open models are closing in on top closed ones fast. The big twist: they’re also pulling other companies into their camp.
Based on reporting by ChinaTalk — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Chinese open models are improving fast, and the source argues that’s changing the fight with closed labs. The latest round of releases — GLM 5.2, K3, and Qwen 3.8 Max — has narrowed the gap with top systems, even though the basic lead held by OpenAI and Anthropic still hasn’t vanished.
The piece’s big claim is that open source has flipped the old “we’ll never catch up” story. Earlier, the pessimistic view was simple: the US has the best compute, the best first-mover advantage, and years of head start in high-end annotation and synthetic data. Chinese models were supposed to lose on every front. But once open models started getting good enough, the contest stopped looking like a one-way race and started looking like a long war.
That shift matters because open models do more than post good benchmark numbers. They create an ecosystem. When K3 triggered pushback in the US, more than 270 tech companies and organizations, including Nvidia and Microsoft, signed an open letter supporting open-source models, and the ban idea went nowhere. The author’s point is blunt: a closed model wouldn’t have gotten that kind of defense from the industry.
Cost is the other lever. The article says Fable 5 is so expensive that even big tech can’t just let everyone use it freely, so expensive models get reserved for a few hard jobs while cheaper ones handle the rest. That is where open models win: they become the default for most token demand, while a small number of premium tasks stay with the pricey closed systems.
Qwen 3.8 27B is the poster child here. The source calls it a model that can run on consumer-grade GPUs, says it has reached around Opus 4.6-level performance, and notes that hardware fans are already deploying it on Mac minis, 4090s, and even a modded 3080. The bigger point is not a single model’s score. It’s that intelligence is getting cheap enough to spread, and once that happens, the market stops being about a handful of giant labs and starts looking like a much messier, much wider fight.
My take — AI-written commentary, not fact-checked reporting
Open source keeps winning the unglamorous part of AI: distribution. That’s the bit that matters, not the ceremonial leaderboard at the top. Closed models can still lead, but the industry keeps acting like access is a side issue, which is how you end up with a very expensive crown and everyone else building the actual kingdom.
Read more about this at: ChinaTalk