Jev 1.13 vs GPT-6 Astra: 17 Demos Side by Side

Which Demos Should You Open First?

GPT-6 Astra was released on September 3, followed by Jev on September 15. The demos people shared cover browser games, practical work, visual experiments, navigation tasks, and tools built around other software. These early demos cannot show everything either model can do.

Start with “Deliver the morning paper,” a GPT-6 Astra browser game from builtbysketch. It places the player on a delivery route with traffic, pedestrians, route choices, and scored paper throws. Its 3,442,211 X views make it the most viewed example here, but the interaction itself is the reason to open it: movement, obstacles, decisions, and scoring all contribute to the same playable idea.

For Jev, Matthew Berman’s “Break down competitor ads” is the first practical demonstration to watch. Berman uses Jev to examine hooks, offers, formats, and the fit between an ad and its landing page. With 1,044,431 X views, it is also the most widely seen Jev example in this group. The demo is interesting because it applies one analysis to several connected parts of an advertising campaign.

“Build an interface from a request” offers another useful view of Jev. In michaeltefula’s creator-built demo, a natural-language request becomes a plan for an interface. The important result is the organized plan, which makes this a good starting point for anyone searching for examples of Jev handling a structured work request.

For a visual GPT-6 Astra project, bentossell’s “50 years of devices” turns a collection of devices into a site where visitors can save the ones they had or wanted. Astra image-generated the devices and built the site. That combination makes it a compact example of generated imagery supporting a browsable personal collection.

Which GPT-6 Astra Visual Demos Are Worth Seeing?

“Video explored as a 3D space-time visualization” by LinusEkenstam presents an uploaded video as a dynamic three-dimensional browser view. A timeline and preview frames remain part of the interface. The demo is worth seeing because it changes how someone can move through and inspect a familiar media format.

vib3coded’s “A knight’s camp in 3D” begins with an image and a prompt, then turns them into a browser scene. The knight cooks, trains, and rests within the camp. Rather than presenting a single static result, it shows several activities inside one setting.

MarcDagatan shared “Drive out of a zombie city,” a night-driving game that can also be opened as a public build. Players steer through traffic and zombies. It is a direct demonstration of a creator moving from showing a game to letting other people try its central driving challenge.

“Draw something that can outrun Jimothy” by YutingM84881 uses a different kind of game loop. You draw a challenger, race it against a raccoon, revise the doodle, and run the race again. The drawing is both the player’s input and the thing being tested, so each revision feeds naturally into another attempt.

In another animation experiment, laszlogaal_ shared a version that lets viewers remove non-moving pixels. The idea came from an earlier animation and its original study. It is a focused visual control built around separating movement from the parts of an image that remain still.

CiusRadu’s “Procedural Japanese Building in Houdini” moves from browser experiments into a 3D production workflow. Seventeen reference images become a procedural Japanese-style building model, presented as a rendered multi-story structure. The result connects a defined set of visual references to a reusable modeling approach.

What Do Jev Work Demos Actually Show?

gajananrx’s “Is this startup stupid?” turns startup evaluation into a playful scoring tool. Enter an idea and it scores novelty, money potential, buildability, and appeal. Those separate dimensions make the response easier to inspect than a single overall reaction, while the playful framing keeps the evaluation approachable.

“Compare structured decisions” places Jev and a language model in front of the same structured request. The value of the demo is the side-by-side process: viewers can watch how each works through an identical task. It is the most direct choice for someone specifically looking for a Jev comparison rather than a standalone Jev application.

How Does Jev Handle Games and Navigation?

“Jev plays Doom” shows an AI reacting to game state and following instructions inside Doom. The interest comes from watching decisions unfold against a changing game situation, rather than reading the final answer to a fixed request.

“A race through Wikipedia” gives Jev a different navigation problem. It must choose links that lead toward a target Wikipedia page. Each selected link becomes the next step in the route, making the decision sequence visible through the pages Jev chooses.

What Do GPT-6 Astra Demos Show Beyond Games?

“The world, with a different perspective.” is an interactive map demonstration. Dragging between two map projections reveals how much the world’s shapes change. It makes the comparison understandable through direct movement instead of requiring a written explanation of the projections.

“Build a live voice client for Gemini” is a browser client for interruptible voice conversations with Gemini. It was built from public product documentation. This is a useful example for people looking for an Astra demo centered on a working software client and live conversational interaction.

DrivingBench places frontier language models in a real comma-equipped Toyota and asks them to drive through a cone course one command at a time. A human supervisor remains ready to brake. Among these demos, it provides the clearest example of model commands being applied to a physical driving task with direct human supervision.

Which Model's Demos Should You Start With?

Choose the GPT-6 Astra examples first if you want interactive browser games, visual transformations, 3D work, maps, a voice client, or a physical driving demonstration. “Deliver the morning paper” offers the broadest playable starting point, while the video visualization and map projection demo focus on exploring familiar information in different visual forms.

Start with the Jev demos when your interest is structured analysis and decisions. The competitor-ad breakdown covers a recognizable work task, the interface demo converts a request into a plan, and the startup scorer separates an idea into four evaluation areas.

For game behavior rather than work, Jev’s Doom and Wikipedia demonstrations make its choices visible step by step. Together, the shared demos give Jev a decision-oriented set of examples and GPT-6 Astra a wider mix of games, visual projects, tools, and 3D workflows.

Demos Discussed