Qwen3.8 Max

Qwen3.8 Max is Qwen's flagship model in the Qwen family, released as the most capable checkpoint in the lineup to date and built upon the architectural foundation of Qwen 3.5. The creator describes it as a 2.4-trillion-parameter Mixture-of-Experts model with 95 billion active parameters, scaling the prior generation while introducing hybrid thinking enabled by default. It is served through QwenCloud, with Qwen planning the first open-weights release of a Qwen-Max-class model the week after launch, signaling a deliberate move toward making top-tier weights accessible to the community.

Qualitatively, Qwen positions the model for end-to-end reliability on multi-day coding projects, research workflows, and long-horizon agent tasks, aiming to take a real project from an empty folder to a finished result with minimal human help. Independent reporting describes the official catalog as exposing a very large context window along with Thinking, Function Calling, built-in tools, and Structured Output, giving the model a tool-oriented profile suited to complex, multi-step deliverables. Together, these characteristics make Qwen3.8 Max a strong fit for teams that need a single model to plan, code, and verify sophisticated workflows rather than answer isolated prompts.

302.AIqwen3.8-maxqwen

Open provider page Provider docs

Quick Info

Provider: 302.AI
Model key: qwen3.8-max
Release date: Aug 3, 2026
Last updated: Aug 3, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $2.16
Output token cost: $6.36

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

302.AI

Coverage

Rost Glukhov's Medium analysis (August 2026) confirms that open weights for both Qwen3.8-Max and Qwen3.8-27B have been released on Hugging Face (at Qwen/Qwen3.8-2.4T-A95B and Qwen/Qwen3.8-27B) and ModelScope following the official announcement. The article describes Qwen3.8-Max as a 2.4-trillion-parameter model with su This third-party piece provides independent corroboration of the weight release timeline and confirms developer access channels for Qwen3.8-Max beyond the Alibaba Cloud Model Studio API. The article contextualizes the model's practical accessibility: even an idealized four-bit representation of 2.4 trillion parameters

302.AI

CoverageBenchmark

CodeOxi's explainer confirms Qwen3.8-Max is a 2.44-trillion-parameter sparse Mixture-of-Experts model that activates approximately 95 billion parameters per token, launched by Alibaba on August 3, 2026 after a July 19 preview. The model features a 1 million-token context window with up to 128k output tokens, matching f The article highlights that open weights at this scale differ from open access — a 2.4T-parameter model does not fit on typical workstation or single-rack inference deployments. Alibaba committed to releasing the weights publicly on Hugging Face and ModelScope alongside a much smaller Qwen3.8-27B variant that most team

302.AI

CoverageBenchmark

BenchLM's independent benchmark tracking page (data as of September 10, 2026) provides third-party performance data for Qwen3.8-Max, recording a composite capability score of 71.6 out of 100 and ranking it 10th out of 232 tracked models. The strongest eligible category is Multimodal & Grounded at rank 5, with the model Capability percentile breakdowns from BenchLM show Qwen3.8-Max at the 95th percentile for Agentic, 87th for Coding, 91st for Knowledge, 91st for Multimodal, and 86th for Instruction Following among eligible categories, with category-specific ranks of 9/152 for Agentic, 20/151 for Coding, and 17/183 for Knowledge. The p

302.AI

Coverage

Alibaba officially announced Qwen3.8-Max on August 3, 2026, positioning it as the most powerful model in its Qwen series to date. The model features a 2.4 trillion parameter Sparse Mixture-of-Experts architecture with a hybrid attention mechanism, activating just 95 billion parameters per token while supporting a conte The model operates as a multimodal foundation with native visual intelligence support and is accessible via APIs on Alibaba Cloud Model Studio for global developers. Model weights are scheduled for release the week following the announcement, and the model can also be experienced on QwenWork, Alibaba's all-in-one workp