aiaimodel.cards

Qwen3.8-Max

Language modelOpen weightsCard summarized

A 2.4-trillion-parameter MoE flagship with a major leap in coding and office productivity

Active (latest Max)Source: Qwen3.8-2.4T-A95B model card (Hugging Face)

Use it via API

$1,051 /month
$21.01 per 1K requests
RAG over docs: 12K in · 600 out · 50K requests/mo · 30% cached · change
Code exampleCompare

Where to call it

  • Alibaba Cloud Model Studioqwen3.8-max
  • Self-host (weights)Qwen/Qwen3.8-2.4T-A95B

Supports

  • Image input
  • Structured outputs
  • Tool calling
  • Web search
  • Prompt caching
Full feature support

Good for

  • Long-horizon agentic coding (Terminal Bench 2.1 86.6%, FrontierSWE 73.5%)
  • Professional work in law, finance and other domains (PLawBench 73.2%, PRBench-Finance 58.3%)
  • Research reproduction (PaperBench 93.0%)

Watch out for

Overview

Alibaba · 05Max, front

Qwen3.8-Max

A 2.4-trillion-parameter MoE flagship with a major leap in coding and office productivity

Cost
$2.00 in
Context
1M
Input / output
$2.00 / $6.00
Max output
131K
Knows up to
—
Released
2 Aug 2026

Qwen3.8 is the first Qwen open release of a Max-class model: a 2.4T-parameter sparse MoE (95B active) built on the Qwen3.5 architecture, aimed at coding, professional work, research and long-horizon agent tasks. The open checkpoint is text-only and always thinks; Qwen3.8-Max on the API adds vision, non-thinking mode, 1M default context and built-in tools. model card

qwen3.8-max
Input / output per 1M
$2.00 / $6.00
Context window
1M
Max output
131K
Knows up to
—
Reasoning
Hybrid (thinking on by default; enable_thinking, thinking_budget up to 262K reasoning tokens)
Weights
Open · Qwen3.8-Max License (open weights); API proprietary

About this card. A model card is the maker's own account of a model: what it's for, how it was trained and tested, and where it falls short. This page condenses Qwen Team, Alibaba's Qwen3.8-2.4T-A95B model card (Hugging Face). Every claim carries its source page.