Stop using your most expensive model for every decision. In our Minesweeper experiment (open-source code below), JEV crushed every other model we tested on speed and cost. JEV: 8s / ~$0.003 Fable 5.1: 100s / ~$2.41 Astra: 128s / ~$8.75 Grok 4.6: 158s / ~$0.55 Where would you put... JEV to work in your company?show more

The repo is open source, so you can run your own benchmark and check the results for yourself. Inside, you'll find „Run in sandbox”, so you can play with it in a minute, no digging through config. Run it and share your results 👇

Fun little test. In my experience, Jev is fast but less accurate than SOTA LLMs on the tasks I tried. It also requires framing problems as structured decisions with predefined outputs. It looks promising for fast data labeling or powering dynamic dashboards.

I’d use it to flag PRs that need an extra security review before we hit merge.

In short, how does it work: the model sees the board + a frontier of unknown cells with "mines still needed" per number. Rule: flag only proven mines, reveal only proven-safe cells, guess lowest-probability only when stuck. Jev agent: probabilities, flag ≥0.85, reveal ≤0.07 More info in the repo:

I'd use JEV for ticket classification like urgency, topic, or team. It's fast, consistent, and much cheaper than wasting high-end model tokens on thousands of small daily decisions

That’s a solid use case, especially at scale.

Almost 3000x cheaper and 16x faster than Astra? That’s not a tweak, that’s a total game changer 🚀

Yeah, the numbers are pretty wild.

Wonder if it's fast enough to play Tetris?

Feels like a great fit for a gating layer, deciding which tasks actually need deeper reasoning from more expensive models, or routing each task to the model best suited for it.

That’s a great use case!

how did you measure the speed and cost for each model's performance?

Regression triage. Sorting 100 failed tests into flake / test fix / infra / real bug is a decision, not an essay. JEV does the first pass, the expensive model only sees what's left.

Yep, makes a lot of sense.