AI decision API latency · Oct 2026

Jev vs OpenAI Decisions vs Cloudflare Clef, which answers fastest?

Across six cities and 990 requests, a full yes/no answer typically took 195 ms from Jev, 208 ms from OpenAI's Decisions API and 344 ms from Cloudflare Clef.

Fastest: Jev

At a glance

7–20 October 2026 · six test locations

MetricJevby TypeSafeOpenAI Decisionsby OpenAIClefby Cloudflare
Typical response time195 msFastest208 ms344 ms
Slower response time238 msFastest707 ms997 ms
Fastest citySan Francisco · 99 msSan Francisco · 120 msMontreal · 248 ms
Slowest cityMumbai · 287 msMumbai · 312 msMumbai · 647 ms
Price per 1M input tokens$0.042Cheapest$0.10$0.24
Output tokensPrice not listedNot chargedNot charged
Question typesYes/no, choice, scoreYes/no, choice, scoreYes/no, choice, score
Image input—YesYes, up to 4
Open weights——Yes, Apache-2.0
Model we measuredjev-1.13.0, pinnedgpt-6-luna, public betaclef (27B)

Methodology →

Response times are our measurements. Prices and features come from each provider's own pages, checked 20 October 2026. A dash means the page doesn't say. TypeSafe → · OpenAI → · Cloudflare → · Clef weights →

Show full benchmark details ↓
DetailJevOpenAI DecisionsClef
Modeljev-1.13.0gpt-6-lunaclef
Slower requests, p95 (median city)238 ms707 ms997 ms
Fastest locationSan Francisco · 99 msSan Francisco · 120 msMontreal · 248 ms
Slowest locationMumbai · 287 msMumbai · 312 msMumbai · 647 ms
Days when every city slowed together000
Requests in this window390 over 13 daily runs300 over 10 daily runs300 over 10 daily runs

Full response time, day by day

Daily median full response time. Less is better.
0200400600 ms11 Oct20 Oct366235183
Show the numbers
DayJevOpenAI DecisionsClef
11 October 2026207 ms142 ms334 ms
12 October 2026230 ms182 ms369 ms
13 October 2026201 ms174 ms284 ms
14 October 2026174 ms236 ms369 ms
15 October 2026182 ms176 ms336 ms
16 October 2026179 ms265 ms445 ms
17 October 2026190 ms179 ms472 ms
18 October 2026183 ms182 ms443 ms
19 October 2026186 ms225 ms343 ms
20 October 2026183 ms235 ms366 ms

Does your location matter?

Yes. All three were slowest from Mumbai, and Clef was the big loser there.

Typical full response time by city: the median of every daily request from that city over the window.
0200400600800 msJev from Amsterdam: 229 ms typical full response time; slower requests 274 ms.OpenAI Decisions from Amsterdam: 213 ms typical full response time; slower requests 953 ms.Clef from Amsterdam: 278 ms typical full response time; slower requests 825 ms.AmsterdamJev from San Francisco: 99 ms typical full response time; slower requests 140 ms.OpenAI Decisions from San Francisco: 120 ms typical full response time; slower requests 344 ms.Clef from San Francisco: 283 ms typical full response time; slower requests 516 ms.San FranciscoJev from Montreal: 117 ms typical full response time; slower requests 167 ms.OpenAI Decisions from Montreal: 156 ms typical full response time; slower requests 574 ms.Clef from Montreal: 248 ms typical full response time; slower requests 564 ms.MontrealJev from Singapore: 243 ms typical full response time; slower requests 301 ms.OpenAI Decisions from Singapore: 299 ms typical full response time; slower requests 840 ms.Clef from Singapore: 453 ms typical full response time; slower requests 1509 ms.SingaporeJev from Tokyo: 160 ms typical full response time; slower requests 201 ms.OpenAI Decisions from Tokyo: 202 ms typical full response time; slower requests 310 ms.Clef from Tokyo: 405 ms typical full response time; slower requests 1176 ms.TokyoJev from Mumbai: 287 ms typical full response time; slower requests 342 ms.OpenAI Decisions from Mumbai: 312 ms typical full response time; slower requests 1235 ms.Clef from Mumbai: 647 ms typical full response time; slower requests 1169 ms.Mumbai
  • Jev
  • OpenAI Decisions
  • Clef
Show exact values
CityJevOpenAI DecisionsClef
Amsterdam229 msslower 274 ms213 msslower 953 ms278 msslower 825 ms
San Francisco99 msslower 140 ms120 msslower 344 ms283 msslower 516 ms
Montreal117 msslower 167 ms156 msslower 574 ms248 msslower 564 ms
Singapore243 msslower 301 ms299 msslower 840 ms453 msslower 1.5 s
Tokyo160 msslower 201 ms202 msslower 310 ms405 msslower 1.2 s
Mumbai287 msslower 342 ms312 msslower 1.2 s647 msslower 1.2 s

Bold is the lowest typical time in that city. "Slower" is the time 95% of the daily requests from that city finished within.

If you're choosing

  • Pick Jev for the fastest, steadiest answers. It also has the lowest input price.
  • Already on OpenAI? Its Decisions API is nearly as fast on a typical request and was quicker in Amsterdam, but the first request after a quiet period is slow.
  • Want open weights or to run it yourself? Clef has them, but it was the slowest of the three in every city we test from.

How we test

Each API gets the same yes/no question ("The customer is explicitly asking for a refund") about the same two-sentence support message, five times per city per day. None of them stream, so we time the whole JSON answer on a new connection. Typical and slower (p95) times both come from those daily requests: 13 daily runs for Jev and 10 for OpenAI and Clef, which we added later.

Jev runs jev-1.13.0, pinned. OpenAI's Decisions API is a public beta and its only model, gpt-6-luna, has no dated version, so its results can change without notice. For Cloudflare we measure clef, not clef-flash.

Full testing methodology →

Good to know

  • OpenAI's first request after a quiet spell is the slow one: in our half-hourly status checks it took about a second in Amsterdam and nearly two seconds in Mumbai, while the two requests right after it took 300 to 400 ms in Amsterdam and 150 to 300 ms in Mumbai.
  • These are whole answers, not first tokens, so don't put them next to the streaming chat APIs' numbers.
  • Jev and Clef take the same request format; OpenAI gets the same question as a predicate in its own format.
  • Clef ran on Cloudflare's free Workers AI allowance.
  • This page compares speed only, not whether the answers were right.

How does your API compare?

Test your endpoint from the same 6 cities used here and see your response time next to these numbers.

Run a free latency test

No account required · Takes about 30 seconds.

Explore each API