Research / Comparison
Compare checkpoints in context.
Inspect total and active parameters, context, licenses and preliminary weight memory. These models have not been evaluated or deployed by Aguasto.
Compare curated released checkpoints. These references do not establish tested model quality, hardware fit, commercial permission for a specific use or an Aguasto inference service.
| Reference field | GLM-4.5 | Qwen3-235B-A22B |
|---|---|---|
| Publisher / checkpoint | Z.aizai-org/GLM-4.5 | QwenQwen/Qwen3-235B-A22B |
| Architecture | Mixture of experts; hybrid thinking and non-thinking text model | Causal mixture-of-experts language model; thinking and non-thinking modes |
| Total parameters | 355 billionPublisher-rounded approximate count; total determines weight arithmetic. | 235 billionPublisher-rounded approximate count; total determines weight arithmetic. |
| Activated parameters | 32 billionThe active count is not the resident weight size. A null field is not a verified active count. | 22 billionThe active count is not the resident weight size. A null field is not a verified active count. |
| Modalities | text-input, text-output | text-input, text-output |
| Native context / documented limit | 131,072 tokensCheck notes for native context, optional extensions and configuration-limit distinctions. | 32,768 tokensCheck notes for native context, optional extensions and configuration-limit distinctions. |
| License | MIT (model weights)The checkpoint card identifies the weights as MIT and describes commercial use. The separate code repository has an Apache 2.0 LICENSE. Check the terms for each downloaded artifact; this description is not a deployment clearance. | 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. |
| Download access | Open download — license conditions apply | Open download — license conditions apply |
| BF16 / FP16 weight arithmetic | 710 GB661.239 GiB. Uses the total at 16 bits; excludes cache, activations and overhead. It is not the size of a quantized checkpoint. | 470 GB437.722 GiB. Uses the total at 16 bits; excludes cache, activations and overhead. It is not the size of a quantized checkpoint. |
| Checkpoint precision notes | This checkpoint is BF16; the publisher also lists a separate FP8 checkpoint. FP8 needs compatible hardware and runtime kernels. | The official checkpoint uses BF16. A quantized derivative needs its own artifact, quality evaluation and compatible serving implementation. |
| Deployment notes | Parameter counts are the publisher's rounded, approximate counts; active parameters do not represent the full weight memory footprint. The config specifies 131,072 positions; the card describes this as 128K context. Long context and batching increase cache memory requirements. Framework links are documentation starting points. Aguasto has not evaluated or deployed this checkpoint, and no throughput or GPU-fit result is claimed. | 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. |
| Primary sources | Z.ai GLM-4.5 model card (external source) · GLM-4.5 checkpoint configuration (external source) · GLM-4.5 code repository license (external source) | Qwen3-235B-A22B official model card (external source) · Qwen3-235B-A22B license (external source) · Qwen3-235B-A22B checkpoint configuration (external source) |