Is this a benchmark that crowns Jev or Laya the winner?

No. The author frames it as controlled observations, not a ranking: 53 experiments and 4,630 decisions on specific versions, with Laya's docs noting the base checkpoint needs fine-tuning for such decisions.

Do I need a Jev API key just to run the test suite?

No. The suite has 103 tests covering rules, runner, analysis, and app that run without a key; the key is only needed for live Jev decisions, while Laya weights download from Hugging Face.

Why does Laya cooperate in one chart but defect when options flip?

That is the position-bias control: with reversed option order Laya changed answers in 3 of 10 comparable cases, including cooperating versus defecting, while Jev kept its answer in 9 of 9.

Can I play a game myself instead of only watching replays?

Yes. Run python experiments/app.py for the tabbed browser app with live matches and a play-against-the-models seat, python experiments/viewer.py for a static replay, or run_tournament.py with human as a player.

What does each model actually receive in a match?

Both receive a JSON state and a typed multiple-choice question and return the chosen option, a probability for every option, and a confidence value. Jev is called through its hosted API with a Choice question, while Laya runs locally on PyTorch with open weights from Hugging Face, so only the Jev side needs a key.

How do the Monty Hall win rates line up with theory?

Both match theory closely. Jev, which always switches, won 69.5% of 200 games against a theoretical 66.7%, while Laya, which always stays, won 33.5% against 33.3%. That is evidence the harness plays the game correctly rather than a claim about either model's general skill.

How is the browser app secured, and what can it do?

It uses Python's built-in web server, listens only on 127.0.0.1, runs one match at a time, and requires a secret token embedded in the page so other sites cannot start runs and spend Jev credit. Every form value is validated before it becomes a runner argument, and each tab shows the game, its theory, per-move confidence, replay, and a seat to play yourself.

CommunityJev Demos & Experiments

Jev_vs_Laya_decision_games

Harness pitting Jev against Laya across six game-theory games with identical rules, plus a browser app to watch matches or play against the models yourself.

See highlights

Open github.com

Stars: 1
Language: Python
Published by: Community
Author: minanayernia
Last verified: Oct 2026

minanayernia/Jev_vs_Laya_decision_games

User repository on GitHub

View on GitHub

Jev (TypeSafe AI) vs Laya (Convai Innovations) in six game-theory games: Rock Paper Scissors, Prisoner's Dilemma, Ultimatum, Beauty Contest, Monty Hall and more. Experiments, analysis, and a browser app to watch them play or play against them yourself.

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Two AI decision models, Jev and Laya, play six classic game-theory games under identical conditions: the same rules, the same information and the same questions. The repository contains the game engine, experiments, statistical analysis, and a browser app where you can watch them play, or play against them yourself. The goal is not to crown a winner. It is to see how each model behaves in the same controlled environment, and how that compares with what game theory predicts.

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Python

Highlights

Six games with typed decisions: Prisoner's Dilemma, Matching Pennies, RPS, Ultimatum, Beauty Contest, and Monty Hall.
Jev adapts to history while Laya mostly repeats fixed choices; both answer deterministically without mixing.
Browser app shows per-move confidence and full distributions, with replay and a seat for a human player.
Reproducible runs with config, decisions, rounds, and summary JSON per experiment plus token cost tracking.

Watch out

Findings cover the tested Jev and Laya versions in this setup only; Laya base checkpoint notes near-chance decisions without fine-tuning. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

tev1-50-use-cases

Fifty runnable tev1 decision-model examples through local Ollama, from ticket triage and guardrails to six playable games, with a live request viewer and scoreboard.

Jev Demos & Experiments#community#github

Community

PromptEngineer48/tev1-50-use-casesPython

Cua

29.3k

The Cua computer-use platform documents a jev-use recipe where TypeSafe Jev picks bounded action IDs and Cua Driver executes clicks and fills in the browser.

Jev Demos & Experiments#community#browser#computer-use

Community

trycua/cuaRust

celesto

Sandboxed virtual machines for agents: each sandbox is a microVM that starts in about 500 ms, keeps files and state between sessions, and runs locally or in Celesto Cloud behind a CLI and Python SDK.

Jev Demos & Experiments#community#jevbest

Community

CelestoAI/celestoPython

FluidUse

306

Local macOS computer-use automation on Apple silicon using on-device laya typed decisions and CUA-S1-FORMS for filling, via the Accessibility API.

Jev Demos & Experiments#community#browser#local

Community

FluidInference/FluidUseSwift

Fast Browser Use

221

An agent skill that reproduces Jev-style single-step logits over indexed browser controls using local Qwen weights for offline browser use.

Jev Demos & Experiments#community#browser#open-models

Community

APUS-AI-Lab/fast-browser-usePython

arc-cua

314

A macOS action layer that lets a planner hand a bounded desktop subtask to a fast decision model which runs the UI loop, plus a driver you can run on its own. Jev is the default backend but the transport is pluggable, and the planner lives outside the package.

Jev Demos & Experiments#community#jevbest

Community

shhivv/arc-cuaPython

From the Jev guides

Original write-ups on what this entry is for.

Per-Tick Decisions in Games

Sep 28, 2026

A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.

Jev Demos & Experiments#gaming#agents#control

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai