Model details

Qwen3.8 Max

Qwen3.8-Max is positioned as a flagship release in the Qwen family, introduced through an official blog post titled "Qwen3.8-Max: A New Bar for Coding and Cowork." Independent coverage describes it as a 2.4-trillion-parameter model launched alongside a smaller 27B variant whose open weights were promised for the following week, signaling a continued dual strategy of a large closed API model paired with an open-weight sibling. The announcement drew unusually strong engagement, surfacing as a high-scoring Hacker News thread within a day of publication, which suggests the release resonated with practitioners watching the open-weights frontier.

In practical terms, the model is framed around coding and "cowork" agentic scenarios, framing that reflects the lab's broader shift toward closed API products for advanced workloads while still feeding the open-source ecosystem through companion releases. The third-party commentary situates Qwen3.8-Max as a top-tier closed model that would have led the open-weights category absent a competing release, highlighting its relevance to teams building AI pair programmers, multi-step coding assistants, and collaborative agent pipelines. Developers evaluating it should weigh its scale and coding focus against the open-weight 27B option expected shortly after launch.

Alibaba (China)qwen3.8-maxqwen

Open provider page Provider docs

Quick Info

Provider: Alibaba (China)
Model key: qwen3.8-max
Release date: Aug 3, 2026
Last updated: Aug 3, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $1.77744
Output token cost: $5.33231

Limits

Output tokens: 131,072 tokens
Context window: 1,000,000 tokens

Transparent token rates

Compare Qwen3.8 Max pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Qwen3.8 Max

Alibaba (China)

Coverage

MLQ News confirms the core Qwen3.8-Max architecture — 2.4 trillion total parameters with 95 billion active per inference through sparse mixture-of-experts, and a 1 million token context window — and ties it to concrete comparative benchmarks: a Frontend Code Arena score of 1,668, which trails Anthropic's Claude Opus 5 For developer economics, MLQ News reports Qwen3.8-Max is priced at roughly 24% of Claude Opus 5's output token cost and about 40% of Claude Opus 5's input token cost, a significant pricing delta that materially shifts build-versus-buy math for teams evaluating hosted frontier-class access via Alibaba Cloud Model Studio

Alibaba (China)

Coverage

Alibaba launched Qwen3.8-Max on August 3, 2026 as the largest model in its Qwen family, featuring 2.4 trillion total parameters with a Sparse Mixture-of-Experts architecture that activates only 95 billion parameters at inference, paired with a hybrid attention mechanism designed to lower computational cost and latency The launch bundles several developer-facing capabilities: Qwen3.8-Max can ingest hundred-page documents, full television series, or 100-hour livestreams into searchable knowledge bases, and supports visual tasks such as editing footage, generating animations from text, reconstructing web projects from a single screensh

Alibaba (China)

CoverageBenchmark

innfactory.ai provides a structured version matrix distinguishing the proprietary Qwen3.8-Max API variant from a separate open-weights Qwen3.8-2.4T-A95B release that targets Hugging Face and ModelScope, with native context of 262,144 tokens extensible to roughly 1M. The Max API variant supports a 1M token context windo Licensing diverges materially between variants: Qwen3.8-27B is Apache 2.0 and unrestricted for self-hosting and commercial use, whereas the open-weights 2.4T-A95B ships under a bespoke licence that imposes a separate commercial agreement on MaaS or AI-assistant providers above US$50M in revenue, and unlike the API vari