Qwen/Qwen3.5-397B-A17B model card

Fields

id

sources/hf-qwen3-5-397b-a17b-card

URL

https://huggingface.co/Qwen/Qwen3.5-397B-A17B

Publisher

Qwen Team, Alibaba

Kind

repo

Published

2026-02-16

Accessed

2026-10-18

notes

Fetched 2026-10-18. Weights repo created 2026-02-16 (HF API createdAt 2026-02-16T04:55:12Z, commit 8472618112ab). Card: Qwen3.5-397B-A17B, 397B total and 17B activated, 262,144 native context, Apache 2.0, vision-language. The benchmark table is a vendor claim: MMLU-Pro 87.8, GPQA 88.4, LiveCodeBench v6 83.6, AIME26 91.3, SWE-bench Verified 76.4, Terminal Bench 2 52.5. The same table puts K2.5 at MMLU-Pro 87.1, GPQA 87.6 and SWE-bench Verified 76.8. The card says the artifacts are compatible with Hugging Face Transformers, vLLM, SGLang and KTransformers. Weights are BF16, no quantization_config. HF safetensors metadata reports 403,397,928,944 parameters, almost all BF16 (403,397,920,304) plus 8,640 F32.

Notes

Fetched 2026-10-18. Weights repo created 2026-02-16 (HF API createdAt 2026-02-16T04:55:12Z, commit 8472618112ab). Card: Qwen3.5-397B-A17B, 397B total and 17B activated, 262,144 native context, Apache 2.0, vision-language. The benchmark table is a vendor claim: MMLU-Pro 87.8, GPQA 88.4, LiveCodeBench v6 83.6, AIME26 91.3, SWE-bench Verified 76.4, Terminal Bench 2 52.5. The same table puts K2.5 at MMLU-Pro 87.1, GPQA 87.6 and SWE-bench Verified 76.8. The card says the artifacts are compatible with Hugging Face Transformers, vLLM, SGLang and KTransformers. Weights are BF16, no quantization_config. HF safetensors metadata reports 403,397,928,944 parameters, almost all BF16 (403,397,920,304) plus 8,640 F32.

Cross-references

This record references no other records.