Jev vs layahost at a glance
| Jev (TypeSafe AI) | layahost | |
|---|---|---|
| API | /v1/systemone, /v1/models | Same endpoints and shapes, plus the simpler /v1/decide with templates |
| SDKs | Official Python and JavaScript | The same official SDKs, via TYPESAFE_BASE_URL |
| Model | Jev (proprietary) | Laya (open source, Apache-2.0), English and multilingual checkpoints; or Cloudflare Clef-flash (9B) and Clef (27B), open weights, per request |
| Pricing | $0.042 per million input tokens | Per decision, any text length: plans from $9/month ($4.90 per million on Pro), or $15 per million prepaid. Clef models: $0.10 / $0.50 per million input tokens |
| Free credit | $5 on sign-up | $1 on sign-up (66.7K decisions), plus a free demo on the home page |
| Rate limit | 1,200 requests/min | 600 requests/min per key, raised on request |
| Hosting | US company; server region not published | Laya in the EU (Helsinki, Finland); Clef on Cloudflare Workers AI |
| Your text | See their terms | Processed in memory, never used for training, logged only if you opt in |
Jev figures from its public documentation, September 2026.
When layahost is the better fit
- You want to choose on evidence. Upload your own labelled texts and run Laya, Clef-flash and Clef side by side: accuracy, AUC, latency and cost per 1,000 rows, in a few minutes. See models.
- Longer texts. Jev charges per input token, layahost per decision. On the Pro plan, with one question per request, flat pricing is cheaper above about 117 tokens: support tickets, emails, reviews and documents.
- Decision flows. Chain questions into a flow in a visual editor and run it with one request; only the questions on the path are billed. See decision flows.
- EU data residency. Requests are processed on servers in the EU and the text is not stored by default.
- An open model. Laya is Apache-2.0. If you outgrow a hosted API you can run the same model yourself, with no change in behaviour.
- Ready-made decisions. Moderation, spam, routing, intent, sentiment, priority, prompt injection and language as one-line calls to
/v1/decide.
When to stay with Jev
- Very short texts at high volume. Below about 117 tokens per request, Jev’s per-token price is lower.
- Accuracy is everything and the price per token matters. Independent tests put Jev well ahead of Laya on hard questions. Clef closes much of that gap, but costs more per token here than Jev does. Measure on your own data first.
- You need higher default limits than 600 requests per minute and can’t wait for a raise.
Other decision models, October 2026
Since Jev launched in September 2026, several companies have shipped their own decision models with the same three question types (Liquid AI and OpenAI among them). A few for reference, at the prices their makers publish:
| Model | Maker | Weights | Price per million input tokens |
|---|---|---|---|
| Jev 1.13 | TypeSafe AI | closed | $0.042 |
| Decision-1 | Microsoft | closed | $0.042 |
| Clef-flash / Clef | Cloudflare | open (Apache-2.0) | $0.038 / $0.24 on Workers AI; on layahost, see above |
| Laya | Convai Innovations | open (Apache-2.0) | self-hosted; on layahost, per decision |
Prices from the makers’ pages, Cloudflare’s Workers AI docs and OpenRouter, 14 October 2026. layahost is not affiliated with any of them.
Switch in three steps
- Create an account and verify your email ($1 of credit to test with), then create an API key.
- Set
TYPESAFE_BASE_URL=https://layahost.comandTYPESAFE_API_KEYto your layahost key. - Replay a sample of real traffic and re-check any probability thresholds before moving production.
python · official Jev SDK
from typesafe_sdk import TypeSafeClient
client = TypeSafeClient(base_url="https://layahost.com", api_key="lh_…")
response = client.system_one(
state="I was charged twice this month, please refund one payment.",
questions={"intent": {"type": "choice", "criteria": {"refund": None, "bug": None, "other": None}}},
)
print(response.answers["intent"].choice)FAQ
Is layahost affiliated with TypeSafe AI?
No. layahost is an independent service. “Jev” is TypeSafe AI’s product name, used here only to describe API compatibility.
Will my probabilities be the same?
No. Answers have the same shape, but a different model produces them, so the numbers differ. Treat thresholds as something to re-tune.
Can I use layahost and Jev side by side?
Yes. Because the API is the same, you can route a share of traffic to each and compare answers before deciding.