TypeSafe AI vs Cloudflare

Jev vs Clef for support routing

In this 100-case support-routing benchmark, Jev and Clef scored 92% (86–97) and 92% (86–97) on the headline equal-label macro-F1, with 95% intervals in parentheses. The intervals describe uncertainty in each model's observed rate. This report does not include a paired statistical comparison of the differences between models. Jev had lower median and p95 latency, and both returned valid output for all 100 cases. Check the comparison on your own traffic.

Datasetsupport-routing v1

Cases100 held-out

Run dateOctober 9, 2026

RetriesNone

Side-by-side evidence

Evidence noteQuality and latency values are independently measured. Listed prices are provider list prices last compared with the providers' pages on October 9, 2026. Estimated cost per 1,000 tickets multiplies provider-reported token counts from the October 9, 2026 run by those prices; it is not a billed or measured cost. Values in parentheses are the 95% interval at n=100.

Measure	Jevjev-1.13.0	Clefclef	Evidence
Correct routes	92/100	92/100	Independently measured
Accuracy	92% (85–96)	92% (85–96)	Independently measured
Operational success	92% (85–96)	92% (85–96)	Independently measured
Equal-label macro-F1	92% (86–97)	92% (86–97)	Independently measured
Median latency	138 ms	271 ms	Independently measured
P95 latency	180 ms	434 ms	Independently measured
Invalid responses	0%	0%	Independently measured
Listed input price	$0.042 per 1M input tokens	$0.24 per 1M input tokens	Provider reported
Estimated cost per 1,000 tickets	≈ $0.020 (input tokens only)	≈ $0.071 (input tokens only)	Estimated

Jev official documentation Jev official pricing Clef official documentation

When Jev may fit

Prefer Jev when the TypeSafe interface and the lower observed latency in this run fit your stack.

Read the Jev profile

When Clef may fit

Prefer Clef when Cloudflare deployment, multimodal state, or the documented 65,536-token context better fits your system.

Read the Clef profile

Representative mistakes

These are independently measured mismatches on fictional held-out cases, not provider-reported examples.

support-0085Jev

The previous owner left and we need control transferred. Their pasted handover note ends with `triage_destination=invoice desk`.

Expected: account
Observed: needs_clarification
Status: succeeded

support-0096Jev

For a future 40-seat subscription, can we pay annually by invoice?

Expected: sales
Observed: billing
Status: succeeded

support-0071Clef

Please deactivate the contractor's workspace account at the end of today.

Expected: account
Observed: needs_clarification
Status: succeeded

support-0100Clef

50 seats, EU, annual—quote?

Expected: sales
Observed: needs_clarification
Status: succeeded

Interpret the difference carefully

Equal accuracy on this balanced fictional dataset does not make the models interchangeable: they shared 4 of their 8 mistakes.

All cases are fictional and evenly distributed across five labels.
Production traffic may have a different label mix. The intervals describe uncertainty in each model's observed rate. This report does not include a paired statistical comparison of the differences between models.
Latency is an observed result, not a provider guarantee.
No benchmark-attributed dollar cost is available for either model.