Qwen3-235B-A22B

Qwen3-235B-A22B is the flagship model of the Qwen3 series and one of the largest open language models in the world. It has a total parameter count of 235 billion, with 22 billion parameters activated for each token. This was made possible by an efficient Mixture of Experts (MoE) architecture that includes 128 experts, of which only 8 are engaged at each computational step. Innovative improvements to the attention mechanism ensure high accuracy in context processing and support sequence lengths of up to 128,000 tokens.

One of the key features of Qwen3-235B-A22B is its support for two operational modes: thinking and no-thinking . In thinking mode , the model applies extended logical reasoning and additional computational resources to deeply analyze tasks, achieving the highest level of precision and depth in its responses. The no-thinking mode, on the other hand, is optimized for quickly performing simple tasks such as text formatting, translation, or short-answer queries, without unnecessary use of computing power. This functionality gives users flexibility in balancing speed and output quality.

Qwen3-235B-A22B can be applied in scientific research, software development, test automation, technical documentation processing, and AI agent creation. It is suitable for academic environments, as well as government and corporate projects where high accuracy, scalability, and flexible task customization are essential. Its support for 119 languages also makes it convenient for international use.

Announce Date: 29.04.2025
Parameters: 235B
Experts: 128
Activated: 22B
Context: 131K
Attention Type: Full or Sliding Window Attention
VRAM requirements: 132.9 GB using 4 bits quantization
Developer: Alibaba
Transformers Version: 4.51.0
Ollama Version: 0.6.6
License: Apache 2.0

Use our pre-built public endpoints to test inference and explore Qwen3-235B-A22B capabilities.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.