Total runs:
32.7K
Run Growth:
-2.6K
Growth Rate:
-8.04%
Sponsored by
APIMart
llm-semantic-router / Decision-1.0-Kai
Model's Last Updated: September 22 2026
text-classification
Introduction of Decision-1.0-Kai
Model Details of Decision-1.0-Kai
Decision-1.0-Kai
Kai, from kairos — the right moment to choose.
Your move.
Choose an action, judge a condition, or score against your own rubric. Kai reads the context and candidate descriptions together, then returns a decision with its probability distribution.
572M parameters · 50 languages evaluated · Complete 1K · Fine-tunable
Apache 2.0 · No access request · Try Decision Studio · Decision collection .
Three ways to decide
| Choice | Noul | Score |
|---|---|---|
| Route a support request to the right team. | Check whether the evidence meets a condition. | Rate an answer against an ordered rubric. |
| Dynamic descriptions and probabilities. | Probability of yes, with optional criteria. | Full level distribution and expected score. |
Questions and candidates are supplied at runtime. Your task defines the output vocabulary.
Measured decisions
Higher scores than both Laya baselines in five of six evaluated panels.
Same tasks, complete inputs and released general weights.
| Text language · decision | Kai | Laya English | Laya Multilingual |
|---|---|---|---|
| English · Choice | 41.01 | 40.80 | 30.97 |
| English · Noul (BoolQ)¹ | 71.80 | 72.92 | 62.17 |
| English · Score | 91.80 | 84.71 | 83.06 |
| Non-English · Choice | 48.44 | 39.96 | 43.82 |
| Non-English · Noul | 74.80 | 51.04 | 53.58 |
| Non-English · Score | 87.47 | 82.08 | 83.41 |
Metrics: Choice macro accuracy · Noul balanced accuracy · Score 1 − normalized RPS . Observed point estimates, not statistical significance or wins in every language. ¹BoolQ is a separate 2,303-question common-native supplement; Kai still trails EN. Evaluation details and fixed model revisions .
Evaluated across 50 languages
Kai was evaluated across 50 Belebele languages , alongside causal reasoning, evidence-based Noul and ordered scoring. The five-family comparison contains 53,196 complete-1K decisions , including 41,720 in the common native window. Current coverage and results .
Six-panel benchmark ranking
| Model | Mean ↑ | EN Choice | EN Noul¹ | EN Score | Non-EN Choice | Non-EN Noul | Non-EN Score |
|---|---|---|---|---|---|---|---|
| Nox | 78.57 | 70.82 | 85.24 | 85.00 | 68.47 | 81.78 | 80.10 |
| Sol | 74.76 | 65.96 | 74.93 | 86.63 | 61.41 | 75.34 | 84.29 |
| Kai | 69.22 | 41.01 | 71.80 | 91.80 | 48.44 | 74.80 | 87.47 |
| Laya English | 61.92 | 40.80 | 72.92 | 84.71 | 39.96 | 51.04 | 82.08 |
| Laya Multilingual | 59.50 | 30.97 | 62.17 | 83.06 | 43.82 | 53.58 | 83.41 |
Mean is the equal-weight average of the six panel scores, calculated before rounding. The same complete inputs and metrics are used for every model. Scores and methodology .
Decisions in milliseconds
Measured on an AMD ROCm GPU with the released Kai weights and the complete-input Python API:
| Workload | Median latency |
|---|---|
| One Choice question · 242 tokens | 14.88 ms |
| Eight Choice questions · 242 tokens each | 22.88 ms |
| One Choice question · complete 1K | 18.89 ms |
FP32, batch size up to 8; includes tokenization, packing and device transfers. 10 warmups and 100 timed runs per workload. Request-local input reuse preserves model outputs. Latency curves and measurement details .
Fixed 242 complete tokens per question, with the same synthetic content repeated. Requests above eight questions run in multiple batches. Measurements and CSV .
Optional larger batches:
predict_auto_1k
uses up to 32 same-type questions when padding does not increase. Earlier B8/B32 comparisons showed approximately 23% lower median latency for 32 short questions and 34% for the multi-context fixture; the final guard was validated separately, not timed.
Usage and measurements
.
Make it yours
One state. Many decisions. Use the System One API to submit up to 128 typed questions, or batch the same questions across independent contexts. Results return under your original question IDs.
Supply a state, question and candidates with the Python examples . Continue training on your own hard or soft labels with the included fine-tuning CLI , including checkpoint resume.
The complete 1,024-token budget includes context, instructions, all candidates and special tokens. Overlength requests return an error. Native Choice and Score support 2–255 candidates or ordered levels; the System One and Studio interfaces use 2–10 Score levels. Longer contexts remain under development.
Architecture
Three 22-layer bidirectional encoder paths share multilingual input embeddings. Each decision type has its own interaction layers and candidate readout. Candidate descriptions are scored jointly within a question; multiple questions use batching.
Architecture details · Methods
Built on Vela Encoder . Probabilities are not calibrated confidence, candidate order can affect outputs, and reliability varies by task and language. Attribution and retained third-party terms · License scope .
Runs of llm-semantic-router Decision-1.0-Kai on huggingface.co
0
Total runs
0
24-hour runs
0
3-day runs
0
7-day runs
0
30-day runs
More Information About Decision-1.0-Kai huggingface.co Model
Url of Decision-1.0-Kai
Decision-1.0-Kai huggingface.co Url
Provider of Decision-1.0-Kai huggingface.co
Other API from llm-semantic-router
Updated:March 06 2026
Total runs:
11.8K
Run Growth:
3.5K
Growth Rate:
29.58%
Total runs:
11.1K
Run Growth:
11.1K
Growth Rate:
99.61%
Total runs:
7.9K
Run Growth:
1.8K
Growth Rate:
23.51%
Total runs:
7.3K
Run Growth:
3.5K
Growth Rate:
47.65%
Total runs:
7.2K
Run Growth:
2.0K
Growth Rate:
28.23%
Total runs:
6.2K
Run Growth:
542
Growth Rate:
8.75%
Total runs:
5.4K
Run Growth:
2.2K
Growth Rate:
39.95%
Total runs:
3.1K
Run Growth:
3.1K
Growth Rate:
100.00%
Total runs:
2.6K
Run Growth:
-5.0K
Growth Rate:
-192.44%
Total runs:
2.2K
Run Growth:
147
Growth Rate:
6.71%
Total runs:
1.7K
Run Growth:
236
Growth Rate:
13.58%
Total runs:
1.7K
Run Growth:
1.7K
Growth Rate:
100.00%
Total runs:
1.6K
Run Growth:
-22
Growth Rate:
-1.42%
Total runs:
1.6K
Run Growth:
249
Growth Rate:
16.05%
Total runs:
1.2K
Run Growth:
596
Growth Rate:
48.06%
Total runs:
1.1K
Run Growth:
389
Growth Rate:
34.67%
Total runs:
749
Run Growth:
76
Growth Rate:
10.15%
Total runs:
714
Run Growth:
35
Growth Rate:
4.90%
Total runs:
448
Run Growth:
157
Growth Rate:
35.04%
Total runs:
424
Run Growth:
-96
Growth Rate:
-22.64%
Total runs:
396
Run Growth:
-17
Growth Rate:
-4.29%
Total runs:
325
Run Growth:
-298
Growth Rate:
-91.69%
Total runs:
296
Run Growth:
251
Growth Rate:
84.80%
Total runs:
266
Run Growth:
-210
Growth Rate:
-78.95%
Total runs:
197
Run Growth:
-213
Growth Rate:
-108.12%
Total runs:
153
Run Growth:
-191
Growth Rate:
-124.84%
Total runs:
151
Run Growth:
121
Growth Rate:
80.13%
Total runs:
151
Run Growth:
0
Growth Rate:
0.00%
Total runs:
149
Run Growth:
-203
Growth Rate:
-136.24%
Total runs:
131
Run Growth:
27
Growth Rate:
20.61%
Total runs:
128
Run Growth:
128
Growth Rate:
100.00%
Total runs:
121
Run Growth:
121
Growth Rate:
100.00%
Total runs:
117
Run Growth:
30
Growth Rate:
25.64%
Total runs:
112
Run Growth:
112
Growth Rate:
100.00%
Total runs:
106
Run Growth:
94
Growth Rate:
88.68%
Total runs:
100
Run Growth:
100
Growth Rate:
100.00%
Total runs:
88
Run Growth:
88
Growth Rate:
100.00%
Total runs:
83
Run Growth:
-139
Growth Rate:
-167.47%
Total runs:
82
Run Growth:
-144
Growth Rate:
-175.61%
Total runs:
81
Run Growth:
41
Growth Rate:
50.62%
Total runs:
77
Run Growth:
-82
Growth Rate:
-106.49%
Total runs:
77
Run Growth:
77
Growth Rate:
100.00%
Total runs:
63
Run Growth:
58
Growth Rate:
92.06%
Total runs:
55
Run Growth:
18
Growth Rate:
32.73%
Total runs:
44
Run Growth:
28
Growth Rate:
63.64%
Total runs:
39
Run Growth:
24
Growth Rate:
61.54%
Total runs:
38
Run Growth:
-37
Growth Rate:
-97.37%
Total runs:
31
Run Growth:
17
Growth Rate:
54.84%
Total runs:
30
Run Growth:
-6
Growth Rate:
-20.00%
Total runs:
30
Run Growth:
16
Growth Rate:
53.33%
Total runs:
29
Run Growth:
7
Growth Rate:
24.14%
Total runs:
28
Run Growth:
-222
Growth Rate:
-792.86%
Total runs:
26
Run Growth:
8
Growth Rate:
30.77%
Total runs:
25
Run Growth:
-5
Growth Rate:
-20.00%
Total runs:
25
Run Growth:
1
Growth Rate:
4.00%
Total runs:
23
Run Growth:
7
Growth Rate:
30.43%
Total runs:
23
Run Growth:
8
Growth Rate:
34.78%
Total runs:
23
Run Growth:
-50
Growth Rate:
-217.39%
Total runs:
20
Run Growth:
-22
Growth Rate:
-110.00%
Total runs:
20
Run Growth:
20
Growth Rate:
100.00%
Total runs:
19
Run Growth:
14
Growth Rate:
73.68%
Total runs:
16
Run Growth:
1
Growth Rate:
6.25%
Total runs:
12
Run Growth:
0
Growth Rate:
0.00%
Total runs:
9
Run Growth:
5
Growth Rate:
55.56%
Total runs:
9
Run Growth:
5
Growth Rate:
55.56%
Total runs:
7
Run Growth:
-21
Growth Rate:
-300.00%
Total runs:
7
Run Growth:
3
Growth Rate:
42.86%
Total runs:
4
Run Growth:
-15
Growth Rate:
-375.00%
Total runs:
4
Run Growth:
-6
Growth Rate:
-150.00%
Total runs:
4
Run Growth:
-2
Growth Rate:
-50.00%
Total runs:
4
Run Growth:
1
Growth Rate:
25.00%
Total runs:
4
Run Growth:
-2
Growth Rate:
-50.00%
Total runs:
1
Run Growth:
0
Growth Rate:
0.00%
Total runs:
1
Run Growth:
-1
Growth Rate:
-100.00%
Total runs:
1
Run Growth:
-10
Growth Rate:
-1000.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%