Alibaba’s Qwen 3.8 name covers four materially different models, according to a new OpenRouter guide. The 27B model is downloadable under Apache 2.0, while the 2.4-trillion-parameter A95B mixture-of-experts model uses the Qwen3.8-Max License. Hosted Max has no public weights, and hosted Flash is based on a downloadable Flash-Next model with a separate community license.
The hosted versions also add capabilities that do not map directly to the open releases. OpenRouter says Qwen3.8 Max adds vision input, a mode that can disable reasoning, built-in tools, and a default one-million-token context. Flash similarly packages a long default context and tools around its underlying model. Developers therefore cannot infer deployment rights or features from the family name alone.
OpenRouter routes all four variants and notes that its undated Max alias currently points to the September 2 snapshot. Teams that require reproducibility can use the dated model identifier so a later alias update does not silently change behavior. Pricing and provider context limits vary and can change, so the guide’s figures are a dated catalog snapshot. The durable takeaway is to check the exact model ID, license, and hosted features before choosing a Qwen 3.8 endpoint.