Qwen3.8-Max: specs, pricing and how it performs in a council

Qwen3.8-Max is Alibaba (Qwen)'s model available on Council AI at $2.00 per million input tokens and $6.00 per million output, with a 1,000,000-token context window. Alibaba flagship (released Aug 3, 2026). 2.4T-parameter MoE with ~95B active parameters, 1M context. Multimodal reasoning with visual understanding; the GA successor to Qwen3.8 Max Preview. On Council you do not run it alone: the same prompt goes to up to 10 models from different labs in parallel, and a moderator synthesizes one answer with a consensus score showing where the labs agreed.

Qwen3.8-Max specifications

What Qwen3.8-Max is good at

Alibaba flagship (released Aug 3, 2026). 2.4T-parameter MoE with ~95B active parameters, 1M context. Multimodal reasoning with visual understanding; the GA successor to Qwen3.8 Max Preview.

Council surfaces it for: reasoning, multilingual, complex, vision.

It has native web search, so it can ground answers in current sources rather than training data alone.

Qwen3.8-Max pricing on Council AI

Council does not resell tokens. Every paid plan carries a monthly budget and each model draws against it at cost: $2.00 in / $6.00 out per million tokens for Qwen3.8-Max. Plans start at $19.99/mo (Plus), $39.99/mo (Pro) and $99.99/mo (Ultra, which adds the personal RAG library and MCP server).

If budget is the binding constraint, GLM-5.3 is the cheaper seat in the lineup at $1.40/$4.40.

Using Qwen3.8-Max in a council

A single model — this one included — can be confidently wrong. Council runs Qwen3.8-Max alongside models from other labs on the same prompt, then scores how much they agreed. High consensus is a signal you can act on the answer; low consensus surfaces the specific point they split on, which is usually the part worth checking yourself.

Other Alibaba (Qwen) models in the lineup: Qwen3.7 Flash, Qwen3-Coder.

Frequently asked questions