Alpha
Contact
Support · LLM-jp-4.1 thinking

Will it last?

Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input D ยท 54.4/100

This page is a projection of the one entry record, the Transparency factor that Support covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.

Which domain expands which factor
  • AssessUse & modify + Transparency
  • ImplementData control + Reliability
  • UseReliability
  • SupportTransparency

Common problems & fixes

From the cookbook and cards (ev-cookbook, ev-gguf); not an exhaustive catalogue:

  • Garbled or unparsed output, broken tool calls: You are using the wrong tokenizer or skipping the template. Load the bundled tokenizer, call apply_chat_template, and do not use the openai-harmony tokenizer.
  • Reasoning text mixed into the answer on vLLM: Start the server with the llmjp4 reasoning parser from llm-jp-vllm 0.1.1 or later and vLLM 0.30.0 or later.
  • llama.cpp chat parsing fails: Use upstream v0.6.0 or later and a 4.1 GGUF; the GGUF card still says a fork is needed, so if upstream fails, try the publisher fork named on the card.
  • Special-token text in user input: Input that resembles <|...|> may need custom encoding; parse token IDs instead of decoded text.
  • Wrong whitespace in decoded text: The vLLM example documents that output.text loses whitespace; decode output.token_ids with the model tokenizer.

Versions, changelog & cadence

Hugging Face creation dates (ev-hub-api), the corpus pages (ev-corpus-v41) and the announcement (ev-blog, ev-pcwatch):

DateRelease
2026-03-15 and 16LLM-jp-4 8B base, 8B thinking, 32B-A3B base and thinking
2026-03-30llm-jp-4-8b-instruct
2026-08-14LLM-jp-4 33B base and thinking
2026-09-15 and 16LLM-jp-4.1 8B, 32B-A3B and 33B thinking
2026-09-24official 4.1 GGUF builds
2026-09-284.1 announcement on the NII blog; the repositories were last modified the same day
2026-10-14LLM-jp-4 technical report on arXiv (v1; 33B results to follow)

Revisions are Hugging Face commits; there is no tagged changelog for the weights. The cards link corpus changelogs (for example corpus v4.1 on 2026-03-31 and 2026-04-14). Pin the commit hash.

Community & support channels

The cards give llm-jp(at)nii.ac.jp for questions and a survey form for reporting how the models are used (ev-card-41). Hugging Face discussion tabs exist on the model repositories (none had been opened on the 33B repository on 2026-11-19). The release page lists tools and corpora on GitHub and GitLab (ev-release-page). NII announced a trial API and chat service through the University of Tokyo mdx-MaaS, mainly for researchers (ev-pcwatch). No community forum or commercial support is listed.

Tracked known issues

No public issue tracker exists for the models; these are issues stated in publisher documents:

  • the GGUF card and the cookbook disagree on whether llama.cpp needs a fork (ev-gguf);
  • LLM-jp-4.1 uses a modified Harmony template and is not compatible with the openai-harmony library tokenizer (ev-card-41);
  • original 4.0 GGUF chat templates lack the declaration needed to select the 4.1 parser (ev-cookbook);
  • AIME, GPQA Diamond and long-context tool use remain well behind Qwen3.5 (ev-blog);
  • the models "may still generate inaccurate, inappropriate, or otherwise undesirable outputs" (ev-card-41).

How this scores

The ownership factor this domain covers, drawn from the one entry record.

2

TransparencyDo you know what it is: weights, training, behaviour, and legible terms?

Strong

Corpora, SFT and DPO data and pre-training scripts are public with per-source licences and a technical report, with named exclusions (NWJC, Ceek.jp News, ichikara, AnswerCarefully) and no confirmed 4.1 post-training code (ev-techreport, ev-corpus-v4, ev-sft-dpo, ev-code).

How this scores (AOI sub-dimensions)
Provenance2/5how well we can trace and verify what went into the modelThe canonical source is identifiable (the NII release page links to huggingface.co/llm-jp), weights are safetensors with per-file sha256, the hub scan reports no flagged files and the org publishes official GGUF builds.
Governance3/5how accountable and well-documented the publisher isA national research institute with named authors and a public release record: 4.0 in March 2026, 33B in August, 4.1 in September, technical report in October.
What this means for adoptionYou own the weights, the data lineage and the right to run and change them: Apache-2.0, ungated, with public corpora, SFT and DPO sets and pre-training scripts, so this is one of the most transparent models reviewed. You do not own a deployable safety case: the publisher says the models are for research and development and not for direct production use, offers one self-run Japanese safety benchmark, no guard model and no independent test. Ownership moves to substantial if an independent evaluation lifts safety to 3 or more, and you can reach that yourself by running your own safety and jailbreak tests behind your own guard.

Sources

The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.

Model cardread2026-11-19
The llm-jp-4.1-33b-thinking model card declares Apache-2.0 in the front matter and 11.7T tokens; the 8B and 32B-A3B cards are identical on the points used.
Model cardread2026-11-19
The LLM-jp-4 (4.0) base, 33b-thinking and 8b-instruct cards say "The models released here are in the early stages of our research and development and have not been tuned to ensure outputs align with human intent and safety considerations".
Licenceread2026-11-19
Apache-2.0 is declared in the card front matter and License section of all three 4.1 cards and in section 2.4 of the technical report.
Technical_reportread2026-11-19
The LLM-jp-4 Technical Report (arXiv 2610.19331 v1, 2026-10-14) covers 8B dense and 32B-A3B MoE models trained from scratch on about 12T tokens.
Documentationread2026-11-19
LLM-jp Corpus v4.1 README: English FineWeb-Rescored and MegaMath-Web-Pro-Max-OSS (ODC-BY plus Common Crawl terms) and The Stack v2 code (per-repository licences), hosted on a public GitLab project; copyright stays with original authors; document-removal contact given.
Documentationread2026-11-19
LLM-jp Corpus v4 README lists Japanese, English, Korean, Chinese and code subcorpora with licences (mostly CC BY, ODC-BY, CC BY-SA).
Documentationread2026-11-19
LLM-jp Corpus Midtraining v2 README: 13 English and 2 Japanese sources with licences, including the nemotron_pretraining_code_v2 corpus (491B tokens) under the NVIDIA Data Access Agreement for Model Training and llm-jp-IPT_v0.3.2 under "Various licenses".
Documentationread2026-11-19
The llm-jp-4.1 SFT and DPO dataset cards (ungated) list per-subset licences, and responses were generated with gpt-oss-120b.
Documentationread2026-11-19
github.com/llm-jp/scripts (Apache-2.0) holds LLM-jp-4 pre-training and mid-training configurations (v4-8b, v4-32b which matches the 33B dimensions, v4-32b-a3.8b), installers, corpus-building and checkpoint-conversion scripts.
Documentationread2026-11-19
The Apache-2.0 llm-jp-4-cookbook documents Transformers 5.0.0 or later, vLLM 0.30.0 or later with llm-jp-vllm 0.1.1 or later, and llama.cpp v0.6.0 or later.
Model cardread2026-11-19
The official llm-jp-4.1-33b-thinking-gguf card (modified 2026-10-20) says running with llama.cpp "currently requires a fork of llama.cpp", which contradicts the cookbook; the repositories hold BF16 (66.4 GB) and Q4_K_M (20.4 GB) GGUF files and an imatrix.
Documentationread2026-11-19
The Hugging Face API lists the three 4.1 repositories as ungated, with safetensors BF16 shards carrying LFS sha256, no pickle files, no LICENSE file and a completed hub scan with no flagged files.
Documentationread2026-11-19
The NII release page lists the 4.1 models, corpora and fine-tuning scripts and says "Users of this model are requested to use it appropriately and at their own responsibility, in accordance with" Japan's December 2025 AI R&D and utilisation guideline.
Vendor announcementread2026-11-19
Publisher blog (2026-09-28, Japanese): 4.1 improves post-training for STEM and adds tool calling, the first LLM-jp series with it.
Third-party analysisread2026-11-19
PC Watch (Impress, 2026-09-29) reports that NII released the three LLM-jp-4.1 models on 2026-09-28 and that Apache License 2.0 applies to all.
Documentationunverified2026-11-19
The AnswerCarefully dataset (safety training data and benchmark) is gated (manual approval, licence:other); its LICENSE and README returned HTTP 401, so its terms are documented but unverified.