Skip to content
← All news
4 min read

Alibaba previewed a 2.4 trillion parameter model with zero published benchmarks

Qwen 3.8 launched days after Moonshot's Kimi K3, with Alibaba claiming it trails only Claude Fable 5. No independent or internal benchmark numbers have been published to back that up.

Alibaba claims its new 2.4 trillion parameter model is second only to Fable 5. It has not published a single number.

Alibaba previewed Qwen 3.8 this week, a 2.4 trillion parameter open weight multimodal model, days after Moonshot's Kimi K3 launch. Alibaba says it trails only Claude Fable 5. It has not published a single benchmark number to support that.

What's confirmed

The preview is real and available now through Alibaba's Token Plan, Qoder, and QoderWork at 10 percent of standard pricing, with open weights promised soon. Alibaba says the model should outperform its predecessor on coding, full stack development, data analysis, and office workflows. It is a sparse mixture of experts model, though Alibaba has not disclosed how many of the 2.4 trillion parameters are active per token, a detail that matters a lot for anyone trying to estimate real inference cost.

What's not confirmed

The second-only-to-Fable-5 ranking rests entirely on Alibaba's own internal evaluation. No third party benchmark suite, not Artificial Analysis, not LMArena, has published independent numbers yet. That is the whole story right now: a serious lab claiming a serious result, with the actual evidence still to come.

Why a build studio cares

We evaluate model options on real numbers, not launch-day claims, especially now that two Chinese labs have shipped 2 trillion plus parameter open weight models within days of each other. Qwen 3.8 is worth tracking for the next few weeks specifically to see whether the benchmark claim survives contact with independent testing, before it becomes a real option for a client build.

Next step: see The Decoder's coverage and MarkTechPost's writeup for what's known so far. If you're weighing open weight models for a build and want the current state of play, write to us at hello@gattyworks.com.

AIOpen SourceBenchmarksQwen38AlibabaOpenWeightAIKimiK3ClaudeFable5AIBenchmarksMixtureOfExpertsChinaLLMFrontierAI

Ready to ship?

Send a brief. You will get a written reply, a fixed quote, and a delivery date within 24 hours. If we miss that window, the website fee on your first project is refunded in full.

Average reply under 4 hours. Always inside 24.