Qwen3-235B-A22B

Checkpoint identity

Official identifier

Qwen/Qwen3-235B-A22BQwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source)

Architecture

Causal mixture-of-experts language model; thinking and non-thinking modesQwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source)

Total parameters

235 billionPublished rounded counts; total stored weights determine the preliminary footprint, including all MoE experts.Qwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source)

Active parameters

22 billion per tokenActive parameters are a compute-related count, not the model's full stored weight count.Qwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source)

Modalities

text-input, text-outputQwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source)

Context

32,768 tokensNative configuration where directly verified. Extensions, advertised K-labels and configuration differences are preserved in the notes below.Qwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source)

Release date

Exact release date not independently established in the reviewed artifactsQwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source)

Download access

Public checkpoint referenceQwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source)

License

Apache 2.0The checkpoint includes Apache 2.0 terms, including redistribution license and notice duties, modified-file notices and patent-license conditions. It is not an unconditional guarantee for every commercial use.Qwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source)

Theoretical weight-memory footprint

Using the publisher's approximately 235 billion total parameters at 16 bits per parameter gives 470 GB (437.7 GiB) of nominal weights.

This estimate excludes KV cache, activations, quantization metadata, non-weight buffers and serving overhead. It is not a tested GPU fit. MoE routing activates only part of the network per token while the complete expert weights still need storage.

Quantization

The official checkpoint uses BF16. A quantized derivative needs its own artifact, quality evaluation and compatible serving implementation.

Serving considerations

• Parameter counts are publisher-rounded approximate totals and activated counts; all expert weights still contribute to memory planning.
• The card states 32,768 native context tokens and 131,072 with YaRN. The numeric context field records native context, not the extension; YaRN must be configured and evaluated separately.
• Thinking mode, chat templates and reasoning parsers change token and latency behavior. Aguasto has not tested this model or its framework fit.

Framework documentation

These are third-party documentation starting points. Support depends on the exact checkpoint, runtime version and hardware.