Chinese-Jev
A third-party preprint proposing an approximately 322M-parameter Chinese direct-decision model built on an mmBERT-base encoder.
Chinese-Jev is an experimental research project by independent authors. The paper and code repository are public, but model checkpoints and their license were not verifiably available on the review date.
- Provider
- Chinese-Jev research team
- Reviewed
- October 8, 2026
- Project status
- Experimental
- Verification
- Needs review
- Categories
- research, development
- Compatibility
- Python, PyTorch, mmBERT, Chinese
Inputs
- ✓Chinese text
- ✓Direct-decision questions and candidate options
Outputs
- ✓Paper-described typed decisions
- ✓Candidate probabilities
Interface
- ✓structured-output
Decision tasks
- ✓Classify
- ✓Choose
- ✓Score
Verified model facts
Provider facts and directory notes are separated. Time-sensitive claims include attribution and should be checked again before production use.
Third-party preprint. Chinese-Jev is not affiliated with or endorsed by TypeSafe. All evaluation figures are author-reported, and model checkpoints were not verifiably available on 2026-10-08.
- Current verified version
- Preprint arXiv:2609.36965
- Model family
- Chinese-Jev / mmBERT
- Release date
- 2026-09-29
- Access methods
- research-paper
- Hosted API
- No
- Local deployment
- No
- Downloadable weights
- No
- License
- Unknown
- Supported languages
- Chinese-language tasks described by the paper
Commercial-use note
The repository code is Apache-2.0 and the paper is CC BY 4.0. Model checkpoints were not verifiably released, so their license and commercial-use terms remain unknown.
Context or input limits
The reviewed paper and repository do not publish a production serving context window. No context limit should be inferred from the mmBERT base name.
Pricing
No production API or released checkpoint pricing was verified. The page is a research record, not a purchasable model listing.
Latency and rate limits
No independently reproducible production latency figure was verified in the reviewed sources.
Strengths
- ✓Focuses specifically on Chinese-language direct-decision tasks.
- ✓Publishes a paper and Apache-2.0 research code.
- ✓Uses a comparatively compact paper-reported architecture.
Limitations
- ✓Third-party preprint without completed formal peer review.
- ✓Reported evaluation results come from the authors and were not independently reproduced by Type Safe.
- ✓Model checkpoints and checkpoint licensing were not verifiably available on 2026-10-08.
- ✓The project explicitly states that it is not affiliated with TypeSafe.
Safety notes
- ✓Do not treat paper-reported results as production readiness evidence.
- ✓Wait for verifiable checkpoints and licenses before deployment testing.
Model variants
Validated context is the source-reported tested range. Serving context is the runtime acceptance limit; the two values are not interchangeable.
| Variant | Parameters | Base model | Artifact | Validated context | Serving context |
|---|---|---|---|---|---|
| Chinese-Jev research model | Approximately 322M | mmBERT-base encoder | Paper-reported architecture; checkpoint not verifiably released | Unknown | Unknown |
Implementation guidance
Keep the declared inputs and outputs visible in code review. Validate every model response before it reaches a database, API, or customer-facing action.