← All models

Qwen3.8-2.4T-A95B

Published Publisher Alibabaqwen3.8-2.4t-a95b

Overview

Qwen3.8-2.4T-A95B is an open-weight, post-trained language model in Alibaba's Qwen flagship family. Its sparse mixture-of-experts architecture has 2.4 trillion total parameters and activates approximately 95 billion per token. Qwen announced its availability on August 12, 2026. (github.com)

Capabilities and uses

The model is intended for coding, professional work, research and multi-step agent tasks. It supports adjustable reasoning depth. Alibaba Cloud Model Studio lists function calling, structured output, web search, prefix completion and context caching as supported service capabilities. The weights can also be deployed using compatible inference frameworks. (huggingface.co)

Limits

  • The open-weight model accepts and generates text only; it requires thinking mode and does not support multimodal input. Qwen3.8-Max is a related managed version with additional features, not the same deployment. (huggingface.co)
  • The model card specifies a native context length of 262,144 tokens, extendable to 1,010,000 tokens. Model Studio separately lists a 1,000,000-token context, a maximum output of 131,072 tokens and a maximum input of 983,616 tokens in thinking mode. Limits depend on deployment and configuration. (huggingface.co)
  • Model Studio lists batch inference and fine-tuning as unsupported for this model. The weights are published under the Qwen3.8-Max license; users should review its terms before deployment. (help.aliyun.cn)

Official API pricing

Prices below are Alibaba Cloud Model Studio's published original rates per 1 million tokens, excluding limited-time promotions. They apply to the stated platform and region; the CNY and USD offers are not presented as currency conversions. Checked November 16, 2026.

Official sources