Qwen built something real in the open-source AI world. The smaller open-weight models, the community fine-tunes, the benchmarks that kept showing up on leaderboards — all of it earned Qwen genuine respect from developers who had no obligation to care about a Chinese AI lab. That reputation is now being tested.
Qwen3.6-Plus dropped in the last 48 hours and the reaction on Hacker News has been mixed at best. The flagship model is hosted-only. No open weights. Alibaba is keeping the best model behind its inference API and competing directly with Anthropic and OpenAI in the hosted market. The community that made Qwen relevant is calling it a bait-and-switch.
The frustration is easy to understand.
The Strategy Behind the Pivot
Give away the small models. Use them as free advertising. Lock the flagship behind a paid API. That is the play, and it is not subtle.
Alibaba has been clear about the business logic. Smaller open-weight Qwen variants like Qwen3 32B and 72B continue to be available for download and local use. These models are genuinely good. They run on reasonable hardware. The community has built an entire ecosystem around fine-tuning and deploying them. That ecosystem drives awareness. Developers try Qwen locally, get familiar with the model family, and when they need something more capable, the hosted option is right there.
Qwen3.6-Plus reportedly benchmarks competitively against Opus 4.5 class models. If those numbers hold, Alibaba has a real product for developers who want frontier-level capability without the Anthropic or OpenAI price tag. The business model is coherent.
The problem is the departure from what Qwen represented. Qwen earned its community by being open. Not because of charity — because the open weights strategy worked. Developers adopted the models, published fine-tunes, benchmarked them publicly, and built a reputation that competing labs could not ignore. That reputation is what made Qwen3.6-Plus matter when it launched.
The Comparison Problem
One detail is drawing more heat than it probably deserves in isolation, but it speaks to a pattern people have noticed.
Qwen3.6-Plus benchmarks are compared against Opus 4.5. Not Opus 4.6. The current flagship from Anthropic, the one that actually ships to users today, is not in the comparison. Commenters on HN are calling this out as an attempt to deceive — picking a competitor at a disadvantage rather than fighting the current version.
Is it intentional? Hard to say. Labs cherry-pick benchmarks all the time. But when you are positioning yourself as a frontier model competitor and you skip the most recent version of the model you are comparing against, the optics are bad. Developers notice these things. The people noticing are the same ones Alibaba needs to keep engaged.
What the Open-Source Crowd Is Saying
The HN thread has been running for over 24 hours and the comments are not softening. One commenter pointed out that people are running Qwen3.5 397B on very standard hardware. The community has gotten comfortable with large open-weight models on reasonable equipment. That capability is not going away — the smaller open-weight variants will continue to exist and improve.
But the flagship model, the one that benchmarks competitively with the biggest closed models, will not be among them. Smaller variants are promised at some point, but no specific model size and no timeline. The community is being asked to wait in limbo for open alternatives while the best model sits behind a paywall.
That ask is a different one than Qwen has made before.
The Real Question for Developers
If you are building on Qwen today, this release is worth sitting with.
The immediate question is not whether the model is good. It probably is. The question is what happens when the next flagship model also skips the open-weight release. Vendor lock-in risk is real here. The value of open weights is not just that they are free to download. It is that your application does not break when a company’s strategic priorities shift. Qwen just showed that those priorities can shift.
The open-source models that compete with Qwen3.6-Plus on benchmarks exist and will continue to improve. Gemma 4 posted Arena AI numbers that sit at the top of the leaderboard. Local model quality has reached the point where hosted-only is a choice, not a necessity.
Alibaba is making a bet that developers will pay for convenience. That bet might work. But the community that built Qwen’s reputation is now watching to see whether the relationship they thought they had actually existed.
Sources:
– Qwen Official Blog — Qwen3.6 Announcement
– Hacker News Discussion
