Jev vs synchronous APIs
TypeSafe’s evaluation API is synchronous: you POST state + questions and wait for answers. This page is not “Jev vs HTTP.” It is product architecture: should your API block on that POST?
Unofficial. Latency figures on blogs are vendor or anecdotal — run your own p95. docs.typesafe.ai/api.
Comparison scope
User-facing “classify this comment” can be sync if p95 is acceptable. Checkout-critical paths should not block on a third-party model without a tight timeout and default.
Criteria that decide the architecture
| Axis | Jev (System One) | Synchronous APIs (as architecture) |
|---|---|---|
| Jev transport | Sync POST /v1/systemone |
Your facade may still be async |
| User waits | Only if you block | Your choice |
| Timeouts | Your client | Your client |
| Retries | 429/529 yes; 401/422 no | Same |
Decision quality and control
A sync product API that retries 529 three times can turn a vendor blip into a 10s page. Budget the wait explicitly.
Integration trade-offs
Official SDKs retry 429/529. Set timeouts at the product layer shorter than the SDK’s worst case if the user is waiting.
TypeSafe’s public models page lists jev-1.13 at $0.042 per million input tokens with output tokens free — a vendor claim, not a jev.pro measurement. Confirm on the models page before you bid.
When each approach fits
Prefer Jev when
- The user is already waiting on a judgment and p95 is fine
Prefer Synchronous APIs (as architecture) when
- Your SLA cannot include a model round-trip
Jev is always a sync evaluation. Your API does not have to be.
One hop you may hide
TypeSafe’s evaluation API is synchronous: you POST state + questions and wait. This page is not “Jev vs HTTP.” It is whether your product API should block on that hop. Sketch: user-facing “classify this comment” may wait if your p95 is acceptable → call pinned jev-1.13.0 with a product timeout shorter than the official SDK’s worst-case 429/529 retry budget → default on expiry. Checkout-critical paths should not block on a third-party model. Hide the call behind 202 + job if the SLA cannot include a model round-trip. See vs asynchronous jobs.
A blocking hop still bills at TypeSafe’s vendor-claimed $0.042 / M input (output free). Latency figures on blogs are vendor or anecdotal — we publish no jev.pro SLO. Jaggedness does not change because the user is waiting. System One returns structured answers, not token streams, and does not generate prose while the user waits. Webhooks from TypeSafe: not in the public evaluation reference we used.
How a sync facade turns a 529 into a 10s page
- SDK retries on the user path. Three 529s × backoff is a frozen checkout. Budget the wait explicitly.
- No default. Timeout without a fail-closed action is a hung spinner. Safe default lives in safe defaults.
- Treating Jev as a streaming chat API. No token stream. Do not hold a chunked response open for “tokens.”
- Retry 401/422. Those are not blips. Fix the key or the schema.
- Schema-safe answer as “the request succeeded for the user.” Your 200 is a product claim. A valid Choice can still be the wrong label.
When not to use Jev as a blocking hop
Do not block the user when your SLA cannot include a model round-trip, when checkout or auth is on the path, or when you need streaming tokens. Prefer enqueue + 202. Use a sync product API only when the user is already waiting on a judgment and your p95 is fine.
What this page does not claim
-
No jev.pro latency SLO.
-
Schema-safe output is not the same as factual correctness.
-
No independent bake-off numbers live here.
FAQ
Streaming tokens? System One returns structured answers, not token streams.
Webhooks from TypeSafe? Not in the public evaluation reference we used.
Should the product timeout match the SDK retry budget? Usually no. Product timeout should be shorter if a human is waiting. Let workers use the longer budget.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.
Hub: Comparisons. Siblings: vs asynchronous jobs, api decision endpoint, decision latency. Canonical: https://docs.typesafe.ai.