Laya: Open-Source Counterpart to Jev | Multilingual System 1 Model for Fast AI Decisions
Mohamed Naji Aboo
Overview of Laya as a multilingual System 1 decision model.
Open on YouTubeUnofficial community documentation
Laya AI - Open Source System-1 Decision Model
Laya system one model, compared with Jev. Multilingual, non-autoregressive System 1 decision engine: typed decisions over 100+ languages in a single forward pass.
Jev is a closed API that works out of the box. Laya is open source, runs locally, and is low latency, and it is weak when one choice question has many labels.
Laya vs Jev
Two sources sit in separate tables on purpose. The first is the project's own README. The second is a Hugging Face community dataset that remeasured Laya. Jev cells in both places are published quotes unless a caption says otherwise. The write-up of why the figures disagree is the Laya vs Jev benchmark explained page.
On the official typed-decisions run, the base checkpoints score 0.362 and 0.342, under the 0.461 majority-class baseline. The 0.766 figure belongs to
laya-typed-decisions, fine-tuned on that benchmark's training split. Installing the package does not give you that score.Official self-test
| Jev 1.13.0 | Laya (routed) | ||
|---|---|---|---|
| typed-decisions, 2,000 decisions | 0.727 | 0.766 | +0.039 |
| AG News, 4 labels | 0.910 | 0.950 | +0.040 |
| DAIR Emotion, 6 labels | 0.480 | 0.595 | +0.115 |
| Banking77 (72 vs 77 labels) | 0.870 | 0.425 | Jev leads on >20 options |
| ECE (lower better) | 0.246 | 0.081 | 3× better (post-temperature) |
| p50 latency, 1 question | 236–276 ms | 32.8 ms | 7.8× faster |
| Languages usable | no published benchmark | 45 of 51 | — |
| Weights | closed API | Apache 2.0 | — |
| Cost | $0.042 / 1M tokens | $0 self-hosted | — |
Third-party independent test
| Model | Accuracy | ECE | ms per case |
|---|---|---|---|
| Laya, no fine-tuning | 0.360 | 0.175 | 15.9 |
| Laya fine-tuned on this task | 0.767 | 0.212 | 16.4 |
| Jev 1.13.0 (published) | 0.727 | 0.144 | 710 |
| Teacher self-agreement | 0.735 | — | — |
| Model | Accuracy | ECE | AUROC | Recall | p50 |
|---|---|---|---|---|---|
| Laya, raw | 0.505 | 0.441 | 0.678 | 0.012 | 9 ms |
| Laya, Platt-calibrated | 0.611 | — | 0.679 | — | 9 ms |
| Jev (published) | 0.626 | 0.154 | 0.689 | 0.432 | 239 ms |
| Claude Haiku 4.5 (published) | 0.813 | 0.097 | 0.837 | 0.764 | 687 ms |
| Model | Accuracy | Macro F1 | Top-3 accuracy |
|---|---|---|---|
| Laya English base, same 77-label protocol | 45.91% | 42.90% | 69.42% |
| Cahol fine-tune | 85.55% | 85.53% | 96.43% |
Videos
Community YouTube explainers about Laya and Laya vs Jev. These creators are not affiliated with Convai Innovations or with this site. Prefer the README and BENCHMARKS.md for numbers.
Open Source, Faster Jev is HERE
CoderOne
Walkthrough of the open-weight Laya stack against the closed Jev API.
Open on YouTubeLaya vs Jev: What "7.8x Faster" Actually Means
AICKStudio | Agentic AI
Latency claim from the README comparison, explained for producers.
Open on YouTubeRun This Powerful Jev AI on Your Laptop for $0
Edwin Chen | AI Automation
Local setup angle for the open System 1 model you can run without an API key.
Open on YouTubeFeatures
Fast Inference
Official T4 table: 32.8 ms for one question on laya-multilingual, 39.5 ms on laya. Batched multilingual throughput there is 7.2 ms/question at 10 questions.
Apache 2.0 Open Source
Weights and the pip package are Apache 2.0. Jev, in the same README comparison, is a closed API at a published $0.042 / 1M tokens.
100+ Languages
The README describes laya-multilingual as the 100+ language checkpoint. On the published 51-language MASSIVE sweep it clears 3× random on 45 of 51 languages.
No LLM Hallucinations
The project says there is no text generation, so there is nothing to parse. Outputs are choice, score, and noul. Wrong labels still show up in the tables below.
Run Laya locally
Shortened slice of the README Router quickstart. Hardware, preload pitfalls, and the full script are on Get Started. This site does not run the model.
bash
pip install layapython
import laya
from laya import Router
router = Router(preload=True)
state = {
"from": "user@acme.com",
"subject": "Duplicate charge on invoice #4411",
"body": "Hi, we were billed twice for March. Please refund the duplicate today or we will cancel our plan."
}
questions = {
"department": {
"type": "choice",
"instructions": "Which department should handle this request?",
"criteria": {
"billing": "invoices, payments, refunds",
"technical": "bugs, outages, system errors",
"sales": "pricing, new contracts",
"other": "everything else"
}
}
}
res_en = router.predict(state, questions)
print("Department :", res_en["answers"]["department"]["choice"])