Alibaba’s Qwen3.8-Max Rivals Fable 5, Goes Open

[erek]

“Why It Matters​

Qwen3.8-Max is the third major Chinese frontier model to claim near-parity with a leading US lab in recent weeks, following Kimi K3 and earlier releases from DeepSeek and Zhipu’s GLM 5.2. A confirmed open-weight release next week would let outside researchers and businesses actually test Alibaba’s claims directly, unlike some rivals whose benchmark numbers have gone unverified for weeks after launch.

Alibaba’s performance comparisons are self-reported, and Bloomberg’s account does not cite independent verification from a third-party benchmark platform. Chinese labs making bold launch-day claims that later prove harder to reproduce at scale has been a recurring pattern this year, and Qwen3.8-Max’s numbers deserve the same scrutiny.

One of the most powerful models available today, comparable to leading frontier AI models.
Qwen Team, Alibaba

Bottom Line​

Watch whether Alibaba actually ships the open-weight release next week as promised, and whether independent platforms like Artificial Analysis or LMArena confirm Qwen3.8-Max’s claimed benchmark scores once it’s testable. The gap between launch-day claims and verified performance has been the defining pattern of this year’s Chinese model releases.

For teams evaluating open-weight models, per Relve, an AI tools intelligence platform, the practical move is to wait for independent benchmark confirmation before making any infrastructure decisions around Qwen3.8-Max, the same caution warranted for every rapid-fire China model launch this year.“

Source: https://relvehq.com/blog/noise/alibabas-qwen3-8-max-rivals-fable-5-goes-open

 

[Shoganai]

They're also doing this at one-thousandth the cost and energy consumption.

 

[workshop35]

They're also doing this at one-thousandth the cost and energy consumption.

I'm assuming a lot of the data center buildout is designed around the high cost and electrical requirements of Anthropic and OpenAI. It's going to be interesting if China's cheap open source models throw the capex situation into disarray. It wouldn't be surprising if the US gov ended up banning their use just to protect the billions tied up in research and construction.

 

[M76]

Alibaba’s performance comparisons are self-reported, and Bloomberg’s account does not cite independent verification from a third-party benchmark platform. Chinese labs making bold launch-day claims that later prove harder to reproduce at scale has been a recurring pattern this year, and Qwen3.8-Max’s numbers deserve the same scrutiny.

Not just this year, it always has been the case.

But then again the US ones are also exaggerating their capabilities constantly, just maybe not to the same magnitude.

 

[LukeTbk]

View: https://youtu.be/jz2wF4m7YWE

 

[westrock2000]

Good for them. As you all know 7 out of 10 for reals real life Americans don’t even want computers in America, so no harm no foul I guess.

 

[erek]

View: https://youtu.be/jz2wF4m7YWE

tldr summary ?

opinion?

 

[LukeTbk]

tldr summary ?

opinion?

seem good and seem like it is reasonably priced for the hosted by alibaba monthly option, but it is really not mythos/Fable.... nothing is yet, still a complete on his own tier level of ability to work for a whole day alone. But it is a good 4.8 opus/sonnet competitor....

will have to see the:
https://artificialanalysis.ai/models

They are often token hungry and eat a lot of their cheaper per millions price point. qwen 3.7 max often costing more than gpt 5.6 sol max because of it.