Alibaba’s Qwen3.8-Max: 2.4 Trillion Parameters, No Benchmarks Yet
Alibaba previewed Qwen3.8-Max on July 19, 2026, a 2.4 trillion-parameter multimodal model available at 10% of standard pricing, but without benchmark results.
On July 19, 2026, Alibaba's Qwen team previewed Qwen3.8-Max-Preview at the World AI Conference in Shanghai. The model carries 2.4 trillion parameters, making it the team's first multimodal model to exceed 1 trillion parameters. It is available now through Alibaba's Token Plan at 10% of standard pricing. The catch: no benchmark table, model card, or license has been published. The announcement landed two days after Moonshot AI released Kimi K3, a 2.8 trillion-parameter open-weight rival.
What happened
| Detail | Fact |
|---|---|
| Announcement date | July 19, 2026 |
| Venue | World AI Conference (WAIC), Shanghai |
| Model name | Qwen3.8-Max-Preview |
| Parameter count | 2.4 trillion |
| Preview pricing | 10% of standard Token Plan pricing |
| Open-weight release | Announced as coming soon, not yet available |
| Benchmark table published? | No |
Alibaba’s Qwen team posted the preview announcement on July 19, two days after Moonshot AI shipped Kimi K3. According to the Qwen account, the model is “one of the most powerful available today, comparable to leading frontier systems” and is described as “second only to Fable 5” among the systems it tested internally.
Qwen developer Shuai Bai provided more technical context. He confirmed this is the team’s first multimodal model above 1 trillion parameters, with support for text, images, video, and documents. Alibaba says Qwen3.8-Max should outperform Qwen3.7-Max on coding, full-stack development, data analysis, and office workflows. Those claims rest on internal testing for now.
How does it compare to Kimi K3?
| Model | Developer | Parameters | Open-weight? | Announced |
|---|---|---|---|---|
| Kimi K3 | Moonshot AI | 2.8 trillion | Yes | July 17, 2026 |
| Qwen3.8-Max-Preview | Alibaba | 2.4 trillion | Planned (soon) | July 19, 2026 |
Kimi K3 launched with weights available for download. Qwen3.8-Max is still preview-only, with open-weight access described as coming soon. For teams that need to self-host today, Kimi K3 is the only option of the two. For teams that just want API access at a steep discount, Qwen3.8-Max is live right now. Our earlier coverage of what Kimi K3 means for the market has more background on that launch.
Why it matters
Two multi-trillion-parameter models announced within 48 hours signals that the frontier is moving fast inside Chinese AI labs. Both models sit in a size class that, until recently, was exclusive to closed systems from US labs.
The pricing angle is the most immediately useful fact. At 10% of standard rates, the preview window is a low-cost way to test whether a 2.4 trillion-parameter multimodal model fits a real workload, before committing to full pricing. That matters most for teams working on document processing, mixed-media analysis, or complex coding pipelines.
The missing benchmarks are a genuine limitation. Alibaba’s self-reported claim of being “second only to Fable 5” cannot be verified until third-party results or the full model card appear. Treat every performance figure from this preview as provisional.
Our take
The timing of this announcement looks strategic more than technical. Dropping a 2.4 trillion-parameter preview two days after Kimi K3 is a visibility play, not a ship date. Alibaba confirmed the model exists and is purchasable, which is meaningful. But a preview without a model card or benchmark table is a marketing event first and a product release second.
That said, 10% pricing on a frontier-class multimodal API is worth a real test. If your team has document-heavy workflows or mixed-media analysis tasks, spinning up a small evaluation this week costs almost nothing. If the quality holds up, you have early data before prices normalize.
For businesses thinking about where AI fits into their operations, this is a reminder that capable models are becoming cheaper and more accessible fast. Whether you are evaluating AI integration for your business or just watching the space, the cost curve for high-parameter models is dropping quickly. If you want help scoping what a practical AI evaluation looks like for your stack, reach out to the Lumien team.
What to do about it
- Sign up for Alibaba’s Token Plan and run your most document- or media-heavy prompts against the preview at the 10% rate before full pricing kicks in.
- Compare outputs against Qwen3.7-Max on the same tasks to check whether the claimed improvement on coding and data analysis holds for your use case.
- If you need self-hosted weights today, evaluate Kimi K3 in parallel since it is already open-weight.
- Wait for the official model card and third-party benchmarks before committing to any production integration.
Run a real evaluation now at 10% pricing, and decide on production use only once the benchmark table is public.
Frequently asked questions
What is Qwen3.8-Max and how many parameters does it have?
Qwen3.8-Max-Preview is Alibaba's latest flagship AI model, previewed on July 19, 2026. It has 2.4 trillion parameters and is the Qwen team's first multimodal model above 1 trillion parameters, supporting text, images, video, and documents.
How much does Qwen3.8-Max cost during the preview?
During the preview period, access is available at 10% of standard pricing through Alibaba's Token Plan subscription.
Is Qwen3.8-Max open-weight?
Not yet. Alibaba announced that an open-weight release is coming soon, but as of the July 19, 2026 announcement, only API access via the Token Plan was available.
How does Qwen3.8-Max compare to Kimi K3?
Kimi K3 from Moonshot AI has 2.8 trillion parameters and launched as open-weight on July 17, 2026. Qwen3.8-Max has 2.4 trillion parameters and was previewed on July 19, 2026, with open-weight access still pending.
