Alibaba Opens Access to Qwen3.8-Max
Alibaba released Qwen3.8-Max through its hosted API and QwenWork on August 3, while saying the 2.4-trillion-parameter model's open weights will follow next week. The company describes it as its most capable Qwen model; independent reporting found strong Arena placements but also showed that several Anthropic models ranked higher, so Alibaba's benchmark comparisons remain vendor-reported.
Alibaba released Qwen3.8-Max on August 3 through its hosted API and the QwenWork productivity platform. In its official announcement, the Qwen team said downloadable open weights will follow next week; they were not available when the hosted model launched.
The model uses a mixture-of-experts architecture with 2.4 trillion total parameters and 95 billion active parameters, according to Alibaba. The company presents it as its most capable Qwen model for coding, multimodal work, and long-running agent tasks.
What is available now
Developers can access Qwen3.8-Max through Alibaba's Qwen API, while QwenWork offers a browser-based route for workplace tasks. South China Morning Post reported that QwenWork entered public beta alongside the model and that Qwen3.8-Max supports a context window of up to 1 million tokens.
The official announcement says the model can work across text, images, video, and documents, and can operate software tools over extended tasks. Those are vendor claims. Teams evaluating the model should test the specific workflows, tool permissions, and failure modes that matter to their deployment rather than treating the capability list as independently verified.
Benchmarks need context
Alibaba published benchmark results that place Qwen3.8-Max near Anthropic's Fable 5 on several coding and agent evaluations. The comparison is useful as a starting point, but it was produced by the model vendor.
Independent reporting provides a more qualified picture. Nikkei Asia reported that Qwen3.8-Max placed fifth on Arena's text leaderboard and second on its vision leaderboard after launch. The Verge likewise reported that the text ranking trailed Fable 5 and three Anthropic Opus models. Leaderboard position can change as votes accumulate, and it does not establish performance for a particular production workload.
The open-weight decision comes next
The scheduled weights release could give teams an additional path for private deployment, adaptation, and inference-stack control. Its practical value will depend on the license, hardware requirements, quantization options, and the quality of the released artifacts.
That distinction matters because the August 3 event was a hosted-model launch with an open-weight release promised for the following week. Until the weights and terms are available, claims about local deployment cost or fine-tuning remain prospective. For now, the release expands the set of frontier-scale models available for API evaluation while leaving the self-hosting trade-offs unresolved.
Key Points
- 1Alibaba launched hosted API and QwenWork access to Qwen3.8-Max on August 3 and said downloadable open weights would follow the next week.
- 2The mixture-of-experts model has 2.4 trillion total parameters and 95 billion active parameters, according to Alibaba.
- 3Alibaba's benchmark comparisons are vendor-reported; independent coverage found strong Arena placements but several Anthropic models still ranked higher.
Scoring Rationale
Qwen3.8-Max is a frontier-scale model available through hosted services, with an open-weight release announced for the following week. Its scale, long-context support, and reported leaderboard results are relevant to model evaluation, but vendor benchmarks and prospective self-hosting claims still require independent validation.
Sources
Primary source and supporting public references used for this report.
View 4 more sources
- Alibaba's new Qwen AI model falls short of 'second only to Fable 5' claimasia.nikkei.com
- China's Alibaba takes another swipe at America's AI supremacytheverge.com
- Alibaba's AI model Qwen3.8-Max widely accessible ahead of open-weights releasescmp.com
- Alibaba's open-weight Qwen3.8-Max takes on long-horizon AI tasks with 2.4 trillion parametersthe-decoder.com
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
