Qwen3.8 2.4T A95B

Qwen3.8 2.4T A95B is the flagship of Qwen's 3.8 generation and the largest model the Qwen team has released, built on a sparse Mixture-of-Experts design with 2.4 trillion total parameters. Qwen describes it as the team's first multimodal model above the one-trillion-parameter threshold, positioning it for coding, long-horizon agentic work, and multimodal reasoning. Thinking and reasoning are always enabled at inference time, with effort adjustable across low, high, and xhigh levels so users can trade response speed for deeper deliberation depending on the task.

The model is engineered to handle very large contexts, supporting a 1M-token window with up to 128K output tokens, which makes it well suited to working with entire repositories, thousands of pages of text, or extended multi-step agentic workflows. Its core strengths are aimed at developers and teams building code-generation tools, autonomous agents, and systems that need to reason over both text and images at scale. Alibaba committed to an open-weight release, and the weights are published on Hugging Face under the Qwen/Qwen3.8-2.4T-A95B slug, with a companion FP8 checkpoint also available, though serving at full precision requires substantial infrastructure given the parameter count.

DevPass (LLM Gateway)qwen3.8-2.4t-a95bqwen

Open provider page Provider docs

Quick Info

Provider: DevPass (LLM Gateway)
Model key: qwen3.8-2.4t-a95b
Release date: Aug 12, 2026
Last updated: Aug 12, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $2.00
Output token cost: $6.00

Limits

Output tokens: 131,072 tokens
Context window: 1,010,000 tokens

Transparent token rates

Compare Qwen3.8 2.4T A95B 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 2.4T A95B

DevPass (LLM Gateway)

CoverageBenchmark

Shattered.io's third-party coverage reconstructs the Qwen3.8-Max release timeline, noting that the cloud API debut occurred on August 2, 2026, with the open-weight variant Qwen3.8-2.4T-A95B following on August 13 on Hugging Face and ModelScope. A dense 27B sibling shipped open weights the next day, August 14. The piece The model is described as a 2.4-trillion-parameter sparse mixture-of-experts with 95 billion active parameters, positioned by Qwen for coding, research, and multi-step agent tasks rather than general chat. The article references an 86.6 benchmark figure and positions the release as near-frontier open weights against pr

DevPass (LLM Gateway)

Coverage

NVIDIA's technical blog confirms details of Alibaba's open-weights release of Qwen3.8-2.4T-A95B, a 2.4-trillion-parameter mixture-of-experts model with 95 billion activated parameters per token. The architecture combines full and linear attention, supports up to a 1-million-token context window with 128K output, and is On Day 0 the model achieved over 4,000 tokens per second per GPU and over 350 tokens per second per user on NVIDIA GB300 NVL72 in FP8, with further gains expected from NVFP4 optimizations. Open-source inference recipes are available for SGLang, vLLM, and NVIDIA Dynamo, and the model can also be deployed via a model-fre

DevPass (LLM Gateway)

Coverage

Next AI Model catalogs Qwen3.8-2.4T-A95B as the open-weights release of Qwen 3.8-Max, released August 12, 2026 as the successor to Qwen 3.5. The page notes it scores 40 on the all-round Artificial Analysis Intelligence Index, the same as its cloud twin Qwen 3.8-Max. Pricing is reported at $2 per million input tokens an The page frames the release as the current flagship in the Qwen line, noting a 42-day gap since the last release against a usual cadence of 176 days. The model is described as requiring a commercial license above $50M in annual revenue and ships without image input and non-thinking mode, distinguishing it from the clou