Breaking

TypeSafe tests Jev as a decision gate for marketing workflows, not a copywriter

September 21, 2026 at 15:46 EDT

Foundation Models
AI Agents
Sales & Business

TypeSafe AI's Jev, a model that returns a choice, a score, or a probability instead of a paragraph, is being tested as the gate inside marketing workflows. It can label a search term, score a draft against a checklist, or send an unsure lead to a person. It does not write the ad.

September 15, 2026 · TypeSafe AI

A gate for the work, not a writer of it

Jev returns a choice, a score, or a yes-or-no probability. Marketing teams are testing it to pass, queue, or stop drafts, ads, leads, and agent output before a person signs off.

$0.042

per million input tokens

output tokens, list price

70–500 ms

company latency range

193.6×

faster on TypeSafe's own workflow evals

Same batch, not the same job

1,000 emails: 9 cents against 62 cents

Column height tracks the reported cost. Each block is about 9 cents.

$0.09

Jev
7 rules, about 6 seconds once parallelized

$0.62

GPT-5.6-class
1 classification, about 5 minutes

About 6.9× the cost, and less work on the taller column. The 9-cent figure is the parallel run. Run one at a time, the same seven rules took about 70 seconds for the same 9 cents.

How a marketing workflow uses it

Decide, then let a writer act

1 · Collect

Ads, terms, drafts, leads

→

2 · Judge

One Choice, Score, or Noul

→

3 · Threshold

Apply, queue, or drop

→

4 · Write

A model or a person, then check again

Why operators are trying it

Thousands of labels can cost cents, not tens of dollars
The answer cannot fall outside the options you defined
Confidence can separate auto-apply from human review

What still does not hold

No copy, no explanation, no web lookup
A valid label can still be the wrong one
Budget changes and claims stay with a person

Read the multiple carefully

1.7× on one call. 100.7× when six calls collapse into one.

An independent test found the gap depends on the baseline. TypeSafe's 444.6× cheaper figure is its high-end workflow claim. A pipeline that still has to write was reported only 15.9% faster and 30.1% cheaper. Start on rows your team has already labeled.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.