Stockmark Open Models: Model List and Licenses
About Stockmark
Stockmark is a Japanese company. On its stockmark organization on Hugging Face it releases Japanese models focused on reading business documents.
What It Releases
- Stockmark-2-100B-Instruct (July 2025): according to its model card, a 100B-class (96B) language model built from scratch with a focus on Japanese. It was pre-trained on about 2 trillion tokens (60% English, 30% Japanese, 10% code) and post-trained with SFT and DPO. The context is 32K tokens, and the project was supported by GENIAC of Japan’s Ministry of Economy, Trade and Industry.
- Stockmark-2-VL-100B-beta (May 2025): a beta version that also reads images.
- Document-reading models (2026): Stockmark-DocReasoner based on Qwen2.5-VL-32B (March) and Stockmark-Nemotron-3-Nano-Omni-JapanDocReader based on NVIDIA’s Nemotron-3-Nano-Omni (July), specialized in reading and reasoning over Japanese documents.
- The earlier stockmark-13b and stockmark-100b are also available.
Licenses
They differ by model. Stockmark-2-100B and the earlier models are MIT, DocReasoner is Apache-2.0, VL-100B-beta uses the Qwen license, and JapanDocReader uses NVIDIA’s open model license. Read each model’s license text before use. This summary is not legal advice.
Running Them Locally
- There are no official GGUF builds (as of 2026-09-26); some models have official AWQ or FP8 builds.
- The flagship 100B-class models do not fit on consumer GPUs. For local use, quantizing one of the 32B-class document-reading models is the realistic option.
Sources: the model card of stockmark/Stockmark-2-100B-Instruct; the model list of the stockmark organization on Hugging Face (all as of 2026-09-26).
Our Coverage and Data
Local Model Watch has not published an article on Stockmark’s models yet. The model list below is collected by code from Hugging Face every day. Part of our publisher index; model families link to their family pages.
At a Glance
| Item | Value |
|---|---|
| Hugging Face organization | stockmark |
| Models on Hugging Face | 14 |
| Articles on its own models | 0 |
| Articles on derived models | 0 |
Recent Models on Hugging Face
The publisher lists 14 models in total on Hugging Face. Below are up to 15 of the most recent generative models (text, image, video, audio and so on), the scope of this site, per organization. Classifiers, feature extractors, research components and repositories without a model card are omitted, and the publisher’s own quantized builds and helper variants are folded into the row of the original model. Parameter counts are left blank for repositories that pack weights into integer types, where Hugging Face’s count is not the real parameter count.
→ Scroll horizontally to see all columns
| Released | Model | Task | Parameters | License | Official variants |
|---|---|---|---|---|---|
| 2026-07-15 | stockmark/Stockmark-Nemotron-3-Nano-Omni-JapanDocReader | vision-language | 33.0B | nvidia-open-model-agreement |
FP8 |
| 2026-03-19 | stockmark/Stockmark-DocReasoner-Qwen2.5-VL-32B | vision-language | 33.5B | apache-2.0 |
FP8-dynamic |
| 2025-07-08 | stockmark/Stockmark-2-100B-Instruct | text generation | 96.0B | mit |
— |
| 2025-05-27 | stockmark/Stockmark-2-VL-100B-beta | vision-language | 96.5B | qwen |
— |
| 2025-03-05 | stockmark/Stockmark-2-100B-Instruct-beta | text generation | 96.0B | mit |
AWQ |
| 2024-05-13 | stockmark/stockmark-100b | text generation | 96.2B | mit |
instruct-v0.1 |
| 2023-10-21 | stockmark/stockmark-13b | text generation | 13.2B | mit |
instruct |
| 2023-08-06 | stockmark/gpt-neox-japanese-1.4b | text generation | 1.4B | mit |
— |
| 2023-01-20 | stockmark/bart-base-japanese-news | — | 125M | mit |
— |
From the Hugging Face API. Last changed 2026-09-27 (JST). Collected by code.
Last updated 2026-09-27 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log and the Hugging Face API.