Now No. 1 Domestically in Both Text and Visual Domains
衡宇 投稿 凹非寺
量子位 | 公众号 QbitAI
Alibaba really knows how to keep things quiet. Before anyone noticed, Qwen 3.7 Preview was suddenly released!
The official Arena account (formerly the Large Model Arena) announced the latest results for Qwen3.7-Max-Preview and Qwen3.7-Plus-Preview.
In the text domain, Qwen3.7-Max-Preview ranked No. 13. This also pushed Alibaba (lab) up to No. 6 overall.
It is also now No. 1 among domestic models.
In the visual domain, Qwen3.7-Plus-Preview ranked No. 16, lifting Alibaba to No. 5 overall.
This is also the top-ranked domestic model in that category.
The official Qwen account also commented: “Can’t wait for the release of the Qwen3.7 series!”
Still, hold on a second. Wasn’t Qwen3.6-Max-Preview only released at the end of April?
Qwen3.6-Max is still just a preview version, and now Qwen3.7-Max Preview is already here… It’s a little confusing whether the official release is moving too slowly or the version updates are happening too fast.
Either way, Qwen, when exactly are the official versions of these two models coming out?
I’m about to cry.
When I woke up this morning, Arena had already published the latest performance results for the Qwen3.7 preview versions.
Qwen3.7-Max-Preview ranked No. 13 overall in the text domain.
Ahead of it are the Claude Opus4.6/4.7 series, the Gemini-3.1/3 Pro series, and the GPT-5.4/5.4-High series, among others.
Qwen3.7-Max-Preview is the only Chinese model in Arena’s text-domain rankings to break into the global top 15.
It also shows strength in more granular categories, ranking in the global top 10 in the following areas:
- Math: No. 7
- Expert Prompt: No. 9
- Software/IT: No. 9
- Coding: No. 10
The other model announced at the same time, Qwen3.7-Plus-Preview, is also performing quite well in the visual domain.
It placed No. 18 in this category, pushing Alibaba’s overall ranking up to No. 5.
This is also the only Chinese model to make the rankings in the visual domain.
Arena also released the results of Expert Arena.
In Expert Arena, Qwen3.7-Max-Preview ranked No. 9 in Expert Prompt only.
There was one other Chinese model in this ranking, Xiaomi’s Mimo v2.5 Pro, which came in at No. 7.
Honestly, what more is there to say?
There’s only one thing to say: please release the official version soon! Please open-source it soon! I want to try it!
In the Qwen3 Era, Model Updates Have Clearly Accelerated
One user said they were a little shocked by how Qwen keeps releasing new models. Before you’ve even finished discussing one model, the teaser for the next one is already autoplaying.
This is definitely not an illusion. If you line up Qwen’s release and update timeline, you’ll see that it really is this way.
(Note: this table focuses on the major milestone versions. In reality, there are many more derivative sizes and specialized models, such as Coder, VL, and Omni.)
Looking at this table, Qwen in the early stage (2023–2024) had a fairly standard release pace, with a new major version every 4–6 months or so.
The turning point was the Qwen3 series.
Released in 2025, Qwen3 includes both dense models and MoE models, with parameter sizes ranging from 0.6B to 235B, and supports two modes: complex reasoning and fast response.
Among them, Thinking mode (for complex reasoning) is mainly designed for complex reasoning, decision-making with long chains of thought, and agent tasks. Non-Thinking mode (for fast response) emphasizes low-latency, rapid replies.
The Qwen3.7 preview versions released so far are the latest achievements in this family.
What stands out is that since Qwen3, the pace of model updates has clearly accelerated, and the interval between major versions has been shortened to 2–3 months.
Especially since the beginning of 2026, Qwen has shown activity almost every month, with version updates moving very quickly from 3.5 to 3.6 and then to 3.7.
More often than not, the pattern is to release a Preview version first so the community and developers can try it out, followed later by the official release.
In a sense, this suggests that the Alibaba Qwen team has entered a phase of rapid experimentation and high-frequency releases.
It brings to mind a tweet from Lin Junyang, Qwen’s former lead, when he stepped down:
Qwen folks, just keep moving forward according to the original plan.
When he left, Qwen was in the 3.5 era and, alongside DeepSeek, was a benchmark name in China’s open-source AI scene.
Whether it is continuing the previous strategy or moving forward under new leadership, Qwen3.5 and 3.7 have indeed been getting faster and better, step by step.
In the past, we used to wait years, or at least a year, to see how a new smartphone evolved. Now, in just a few months — sometimes even a few weeks — we’re watching a model go from immature to powerful.
Some are excited, some are anxious, and some keep tuning through the night, all just to keep up with the curve of intelligence pushing relentlessly forward.
Technology waits for no one.
It is like a surging river. We stand on its banks as both witnesses and participants.
References: