Largest Darwin model · Qwen 3.5 397B base + Darwin V9 FFN transplant · 397B MoE (~17B active) · BF16 GPQA Diamond: 90.9 % — pure greedy, single-sample, NO test-time engine
Darwin-397B-JGOS is the largest and highest-scoring member of the Darwin family. Built on Qwen 3.5 397B as the base, it transplants the FFN (expert) strengths of multiple high-performance models through the Darwin V9 platform, producing a 397B-parameter Mixture-of-Experts model with ~17B active parameters per token.
It reaches 90.9 % on GPQA Diamond with pure greedy decoding (single sample) — surpassing Darwin-28B-REASON (89.39 %, achieved with the Darwin-DELPHI test-time engine) without using any test-time engine at all. This is the highest GPQA Diamond score in the Darwin family to date.
Darwin is VIDRAFT's measuring-result-driven reasoning model family — approximately 20 official models plus 400+ community derivatives, ranking among the top open models on GPQA.
| Role | Model | Contribution |
|---|---|---|
| Base | Qwen 3.5 397B (A17B) |
397B Mixture-of-Experts backbone (~17B active). |
| FFN transplant | Darwin V9 platform (proprietary) | Transplants the FFN (expert) strengths of multiple high-performance models onto the base. |
| Result | Darwin-397B-JGOS (this model) |
397B MoE → 90.9 % GPQA Diamond, pure greedy. |
The full Darwin V9 merge recipe — source models, weighting, and density — is proprietary and not disclosed (trade secret).
| Component | Value |
|---|---|
| Architecture | Qwen3_5MoeForConditionalGeneration (Qwen 3.5 generation MoE) |
| Parameters | ~397 B total / ~17 B active (Mixture-of-Experts) |
| Base | Qwen 3.5 397B (A17B) |
| Precision | bfloat16 |
| License | other |
Darwin V9 transplants the FFN (expert) strengths of multiple high-performance models onto a Qwen 3.5 397B MoE base, then applies trust-weighted evolutionary merging.
The source models, merge weights, and density schedule are proprietary and constitute a trade secret; they are not published.
GPQA Diamond is a 198-question, PhD-level graduate science reasoning benchmark.
| Model | Engine | Accuracy |
|---|---|---|
| Darwin-28B-Opus | Standard | 88.89 % (176 / 198) |
| Darwin-28B-REASON | Darwin-DELPHI (test-time) | 89.39 % (177 / 198) |
| Darwin-397B-JGOS | Greedy (single-sample, no engine) | 🥇 90.9 % (180 / 198) |
Reproducible evaluation settings:
max_model_len = 18432Darwin-397B-JGOS achieves the family's top GPQA Diamond score using nothing but greedy decoding — no Darwin-DELPHI, no majority voting.
vllm serve FINAL-Bench/Darwin-397B-JGOS --tensor-parallel-size 2 --pipeline-parallel-size 3 --dtype bfloat16 --trust-remote-code
@misc{darwin397b_jgos_2026,
title = {Darwin-397B-JGOS: Darwin V9 Platform FFN Transplant on a 397B MoE Base},
author = {FINAL-Bench / Darwin Research Team},
year = {2026},
howpublished = {https://huggingface.co/FINAL-Bench/Darwin-397B-JGOS},
note = {Darwin V9 - 90.9 percent GPQA Diamond (greedy, single-sample)}
}
Darwin-397B-JGOS · Darwin V9 Platform · 90.9 % GPQA Diamond (pure greedy) · FINAL-Bench
model.safetensors-00093-of-00094.safetensors 9.7 GBmodel.safetensors-00092-of-00094.safetensors 9.7 GBmodel.safetensors-00091-of-00094.safetensors 9.6 GBmodel.safetensors-00061-of-00094.safetensors 8.6 GBmodel.safetensors-00062-of-00094.safetensors 8.6 GBmodel.safetensors-00063-of-00094.safetensors 8.6 GBmodel.safetensors-00064-of-00094.safetensors 8.6 GBmodel.safetensors-00065-of-00094.safetensors 8.6 GBmodel.safetensors-00066-of-00094.safetensors 8.6 GBmodel.safetensors-00067-of-00094.safetensors 8.6 GBmodel.safetensors-00068-of-00094.safetensors 8.6 GBmodel.safetensors-00069-of-00094.safetensors 8.6 GBmodel.safetensors-00070-of-00094.safetensors 8.6 GBmodel.safetensors-00071-of-00094.safetensors 8.6 GBmodel.safetensors-00072-of-00094.safetensors 8.6 GBmodel.safetensors-00073-of-00094.safetensors 8.6 GBmodel.safetensors-00074-of-00094.safetensors 8.6 GBmodel.safetensors-00075-of-00094.safetensors 8.6 GBmodel.safetensors-00076-of-00094.safetensors 8.6 GBmodel.safetensors-00077-of-00094.safetensors 8.6 GBmodel.safetensors-00078-of-00094.safetensors 8.6 GBmodel.safetensors-00079-of-00094.safetensors 8.6 GBmodel.safetensors-00080-of-00094.safetensors 8.6 GBmodel.safetensors-00081-of-00094.safetensors 8.6 GBmodel.safetensors-00082-of-00094.safetensors 8.6 GBmodel.safetensors-00083-of-00094.safetensors 8.6 GBmodel.safetensors-00084-of-00094.safetensors 8.6 GBmodel.safetensors-00085-of-00094.safetensors 8.6 GBmodel.safetensors-00086-of-00094.safetensors 8.6 GBmodel.safetensors-00087-of-00094.safetensors 8.6 GBmodel.safetensors-00088-of-00094.safetensors 8.6 GBmodel.safetensors-00089-of-00094.safetensors 8.6 GBmodel.safetensors-00090-of-00094.safetensors 8.6 GBmodel.safetensors-00001-of-00094.safetensors 8.6 GBmodel.safetensors-00002-of-00094.safetensors 8.6 GBmodel.safetensors-00003-of-00094.safetensors 8.6 GBmodel.safetensors-00004-of-00094.safetensors 8.6 GBmodel.safetensors-00005-of-00094.safetensors 8.6 GBmodel.safetensors-00006-of-00094.safetensors 8.6 GBmodel.safetensors-00007-of-00094.safetensors 8.6 GBmodel.safetensors-00008-of-00094.safetensors 8.6 GBmodel.safetensors-00009-of-00094.safetensors 8.6 GBmodel.safetensors-00010-of-00094.safetensors 8.6 GBmodel.safetensors-00011-of-00094.safetensors 8.6 GBmodel.safetensors-00012-of-00094.safetensors 8.6 GBmodel.safetensors-00013-of-00094.safetensors 8.6 GBmodel.safetensors-00014-of-00094.safetensors 8.6 GBmodel.safetensors-00015-of-00094.safetensors 8.6 GBmodel.safetensors-00016-of-00094.safetensors 8.6 GBmodel.safetensors-00017-of-00094.safetensors 8.6 GBmodel.safetensors-00018-of-00094.safetensors 8.6 GBmodel.safetensors-00019-of-00094.safetensors 8.6 GBmodel.safetensors-00020-of-00094.safetensors 8.6 GBmodel.safetensors-00021-of-00094.safetensors 8.6 GBmodel.safetensors-00022-of-00094.safetensors 8.6 GBmodel.safetensors-00023-of-00094.safetensors 8.6 GBmodel.safetensors-00024-of-00094.safetensors 8.6 GBmodel.safetensors-00025-of-00094.safetensors 8.6 GBmodel.safetensors-00026-of-00094.safetensors 8.6 GBmodel.safetensors-00027-of-00094.safetensors 8.6 GBmodel.safetensors-00028-of-00094.safetensors 8.6 GBmodel.safetensors-00029-of-00094.safetensors 8.6 GBmodel.safetensors-00030-of-00094.safetensors 8.6 GBmodel.safetensors-00031-of-00094.safetensors 8.6 GBmodel.safetensors-00032-of-00094.safetensors 8.6 GBmodel.safetensors-00033-of-00094.safetensors 8.6 GBmodel.safetensors-00034-of-00094.safetensors 8.6 GBmodel.safetensors-00035-of-00094.safetensors 8.6 GBmodel.safetensors-00036-of-00094.safetensors 8.6 GBmodel.safetensors-00037-of-00094.safetensors 8.6 GBmodel.safetensors-00038-of-00094.safetensors 8.6 GBmodel.safetensors-00039-of-00094.safetensors 8.6 GBmodel.safetensors-00040-of-00094.safetensors 8.6 GBmodel.safetensors-00041-of-00094.safetensors 8.6 GBmodel.safetensors-00042-of-00094.safetensors 8.6 GBmodel.safetensors-00043-of-00094.safetensors 8.6 GBmodel.safetensors-00044-of-00094.safetensors 8.6 GBmodel.safetensors-00045-of-00094.safetensors 8.6 GBmodel.safetensors-00046-of-00094.safetensors 8.6 GBmodel.safetensors-00047-of-00094.safetensors 8.6 GBmodel.safetensors-00048-of-00094.safetensors 8.6 GBmodel.safetensors-00049-of-00094.safetensors 8.6 GBmodel.safetensors-00050-of-00094.safetensors 8.6 GBmodel.safetensors-00051-of-00094.safetensors 8.6 GBmodel.safetensors-00052-of-00094.safetensors 8.6 GBmodel.safetensors-00053-of-00094.safetensors 8.6 GBmodel.safetensors-00054-of-00094.safetensors 8.6 GBmodel.safetensors-00055-of-00094.safetensors 8.6 GBmodel.safetensors-00056-of-00094.safetensors 8.6 GBmodel.safetensors-00057-of-00094.safetensors 8.6 GBmodel.safetensors-00058-of-00094.safetensors 8.6 GBmodel.safetensors-00059-of-00094.safetensors 8.6 GBmodel.safetensors-00060-of-00094.safetensors 8.6 GBmodel.safetensors-00094-of-00094.safetensors 4.7 GBHave the files? Submit a torrent.
Submit a torrentHave the complete files and permission to share them? Replace the example folder path. The script checks the files, makes a torrent and submits its magnet and checksums. Keep your torrent client seeding. Get your community key from your settings, and read the script before running it.
curl -fsSL https://pirateface.co/package.sh | bash -s -- --repo FINAL-Bench/Darwin-397B-JGOS ./model-folderQwen/Qwen3.5-397B-A17B