Jurin AI
2B
JJ-1-2B
Balanced size. A compact model with a 9.3 ms 3-question latency.
- Model ID
- JJ-1-2B
- Backbone
- Qwen3.5-2B
- Release
- Beta
- Languages
- Japanese & English, text / JSON
61.63%
Japanese accuracy
68.22%
Full-set accuracy
9.3ms
3-question latency
JJ-1
A family of models optimized for enterprise business workflows and speed
Give JJ-1 the context and your questions. It returns probabilities for predefined answers, ready for your application to use. Four sizes, from 0.8B to 9B, built on Qwen3.5, with a Jev-compatible decision interface. The current models specialize in Japanese business workflows, including messages that mix Japanese and English.
Sign in or sign up
Sign in with your Jurin account. Your API key is created the first time you sign in.
Jurin AI
Breaking the message into signals and intentā¦
JJ-1 model family
Flagship models
4B
JJ-1-4B
Higher accuracy. A larger model with 74.45% full-set accuracy.
- Model ID
- JJ-1-4B
- Backbone
- Qwen3.5-4B
- Release
- Beta
- Languages
- Japanese & English, text / JSON
69.60%
Japanese accuracy
74.45%
Full-set accuracy
16.2ms
3-question latency
9B
JJ-1-9B
Our largest model. A 30 ms 3-question latency.
- Model ID
- JJ-1-9B
- Backbone
- Qwen3.5-9B
- Release
- Beta
- Languages
- Japanese & English, text / JSON
74.06%
Japanese accuracy
77.82%
Full-set accuracy
30ms
3-question latency
Model family
Four sizes, one interface
Explore the checkpoints, results and model details. From the smallest footprint at 0.8B to our largest model at 9B. All models support Japanese & English Ā· text / JSON.
All latency figures use 3 questions.
| Model | Backbone | Status | Japanese accuracy | Full-set accuracy | Latency |
|---|---|---|---|---|---|
JJ-1-0.8B Smallest footprint | Qwen3.5-0.8B | Beta | 56.46% | 65.13% | 7.8 ms |
JJ-1-2B Balanced size | Qwen3.5-2B | Beta | 61.63% | 68.22% | 9.3 ms |
JJ-1-4B Higher accuracy | Qwen3.5-4B | Beta | 69.60% | 74.45% | 16.2 ms |
JJ-1-9B Largest model | Qwen3.5-9B | Beta | 74.06% | 77.82% | 30 ms |