Jev Use Cases: 10 Builds Checked, and jevgrep's Real Saving Is 28.6%
By Matt Cullerton, founder of Mavric and Apso
Jev is a decision model from TypeSafe AI that returns choices, scores and probabilities instead of text, at a fraction of LLM cost. A TikTok from Ben Broch (@brochbuilds) shows ten things people built with it in its first weeks. The builds are real, but the headline claim that one of them "cuts coding agent costs by 40%" is out of date: its author now reports 28.6%.
What is Jev?
Jev is built by TypeSafe AI, a San Francisco company that describes it as one of its "System One Models, to be natively used by machines" and pitches "Decisions, not strings." The launch post calls it "a new class of frontier models built to make fast, structured decisions that software can use directly."
The documentation lists three primitives instead of a chat endpoint:
- Choice picks an option from a list and returns choice, probabilities and confidence.
- Score rates a state on a rubric and returns score, probabilities and confidence.
- Noul answers "Is this statement true?" with a value from 0 to 1.
The video's transcript renders the name as "Jeff," which is a speech-to-text error.
When did Jev come out?
The video says Jev "has been out for two weeks." Wikipedia records that it "was released in limited early access on September 15, 2026," alongside a $40 million seed round led by DCVC at a $200 million valuation. That is 19 days before the October 4 video. TypeSafe's public introduction post is dated September 28, 2026, 6 days before.
Adoption was fast. Vercel posted that "Jev was adopted faster than any other model in AI Gateway history. In the first day, @typesafeai reached ~13% of teams, 2x the GPT-5.6 family and 6x Fable 5.1."
What does Jev cost?
TypeSafe lists input at $0.042 per million tokens ($42 per billion) and output as "FREE (too cheap to meter)." Its launch post says "End-to-end response time is 70ms-500ms."
TypeSafe's homepage advertises "193.6x Faster, 444.6x Cheaper" than LLMs on comparable workflows. Wikipedia notes TypeSafe's own caveat: those workflows "were created by members of its model-capabilities team," and the company describes the gains as likely "at the high end of real-world results." The video creator's own quickstart reports "642 ms per call, measured," above TypeSafe's 500 ms ceiling.
Jev is a paid, hosted API behind an early-access waitlist, so every build below adds a per-call bill.
Does jevgrep really cut coding agent costs by 40%?
Not anymore. The card in the video, labeled "Creator's original benchmark · @dzhng," shows Codex alone at $7.62 against Codex plus jevgrep at $4.52, a 40% cut. Its fine print says "Jev cost excluded" and "Solves: 7/10 with jg vs 8/10 baseline," so the saving came with one fewer task solved.
The current jevgrep repository reports $7.62 against $5.44, "a measured 28.6% reduction, rounded to ~30%," on ten tuned Python SWE-bench tasks, with 8 of 10 solved by both. Counting Jev's own API cost, total cost fell 25.8%.
The author explained the change: "it is now intelligence parity with codex/claude's own subagents but with 30% less cost. Yes it's less than the original's 40%, but I traded more intelligence for a bit more cost." A later release post reported Jev's own cost down 59% while "the per task cost stayed the same."
Which Jev builds can you run yourself?
Voice browser. moritzkremb/jev-voice-browser lets you "Control a real browser by voice," with Jev deciding intent and Playwright acting. The repo reports latency of "≈ 330 ms (p50 ≈ 300 ms)" and roughly $0.0002 per call on model jev-1.13.0. Its listed limits: the Web Speech API works only in Chrome and Edge and sends audio to Google, the element snapshot caps at 100 items, one action runs per utterance, and iframes and sites with heavy bot protection do not work.
Postgres queries in plain language. pgjev is a PostgreSQL extension under the PostgreSQL license that adds jev(), jev_prob(), jev_choice() and jev_score() as SQL functions and batches up to 40 rows per request. It quotes about 3.5 seconds and roughly $0.012 for a first pass over 2,000 rows, and about 50 milliseconds on cached runs. The video's example, 6,129 input tokens for about $0.0003, matches the $0.042 per million price.
Call-center routing. A Jev Switchboard demo built with LiveKit has Jev judge intent, frustration and whether the caller wants a human, with plain code rules deciding whether to route or escalate.
The other builds in the video (a 3D file organizer, an inbox ranked by sleep data, a self-filling signup form and an end-to-end store test) carry no attribution, so they are not traced here.
The "100+ use cases" list
The video offers "a list of over 100 Jev use cases." The page it shows counts "133 resources": 27 demos, 71 repos and 35 guides, updated September 24, 2026. Those are links in three categories, and 71 of them are repositories.
Bottom line
Jev is a fast, cheap decision model with several runnable community projects, including pgjev and the voice browser. jevgrep's real saving is about 29% at equal task success, or 26% once you pay for Jev. Treat TypeSafe's 193.6x and 444.6x figures as best cases from its own test workflows.
Sources
- TypeSafe AI home page - pricing and speed claims
- Introducing System One Models & Jev - launch post and latency band
- TypeSafe AI documentation - Choice, Score and Noul primitives
- Jev (AI model) on Wikipedia - release date, funding and benchmark caveat
- dzhng/jevgrep - current 28.6% benchmark
- dzhng on jevgrep v0.4 - why the figure dropped from 40%
- dzhng on jevgrep 0.5 - Jev cost down 59%
- moritzkremb/jev-voice-browser - voice browser repo and limits
- pgjev - PostgreSQL extension
- Vercel on Jev adoption - AI Gateway adoption numbers
- Startup Fortune on Jev adoption - secondary reporting
Published October 4, 2026. Based on a TikTok by @brochbuilds. Full video notes (markdown)