Key Notes
- Anthropic engineer Felix Rieseberg reports that his Saltline shooter took eight hours without human feedback, with token usage he estimates at about $550 under API pricing.
- Matt Shumer describes his Earth simulation as a work in progress, while Peter Yang’s kart-racer walkthrough documents an initial two-round build followed by three additional polishing rounds.
- The projects demonstrate broader AI game-prototyping capabilities, but their creators’ accounts do not establish that a few prompts reliably produce a finished AAA game.
A first-person shooter, an explorable virtual Earth and a cartoon kart racer are among the projects creators are sharing after Anthropic released Claude Fable 5.2. The examples turn the model’s coding claims into something people can try, with game environments, sound and interactive mechanics assembled through AI-assisted development.
The most closely documented shooter comes from Anthropic engineer Felix Rieseberg, who reported an eight-hour run without human feedback and estimated its usage at about $550 under API pricing. Other projects show the same model tackling a much larger world simulation and a more familiar browser-racing game. They are early demonstrations of game production, rather than evidence that a few prompts reliably produce a finished AAA release.
The eight-hour shooter behind the $550 claim
Rieseberg published Saltline as a Claude artifact. He says its 3D assets, audio and textures were created from scratch, with roughly 2.7 million output tokens and 445 million input tokens processed during the build. Much of that input was cached, which matters when interpreting his API-price estimate.
His preview shows a first-person weapon, an industrial environment and an objective to disable three uplink relays. The reported eight hours also put the single-prompt framing in perspective: one initial instruction can set off an extended agent run. The development time and cost are Rieseberg’s figures, and his role at Anthropic makes this a demonstration by someone inside the company.
A virtual Earth that remains a work in progress
Matt Shumer took a different route. In his first-hand review, he describes Fable 5.2 building a voxel engine with destruction physics, then extending that work into an Earth-scale mirror world. A public Earth preview lets readers explore the project.
Shumer says the aim is to reflect real locations and their current weather, with ideas such as New York subway activity tied to actual schedules. He explicitly describes the world as still in progress. The claim is an ambitious, evolving simulation; the preview does not establish that every location or physical interaction accurately reproduces the real Earth.
The kart racer needed iteration, too
Peter Yang’s walkthrough, embedded above, shows how he built a 3D browser racer using Fable 5.2. He reports 62 minutes for the initial two-round result: first asking for the game, then asking the model to critique and improve its own work. He subsequently requested three more polishing rounds before sharing the final build.
The result, Turbo Tails, features a fox racer, jumps, drifting and deployable traps. Its current interface lists three courses, Grand Prix and multiplayer-room controls. Yang’s earlier video treats multiplayer as a possible addition, so the published build’s room options should be distinguished from what that initial walkthrough actually demonstrates.
Together, the projects suggest that AI game prototyping is moving beyond static scenes toward systems that combine graphics, movement, sound and rules. That can shorten the path from an idea to a playable experiment. It still leaves developers to judge whether the game is enjoyable, reliable and ready for a wider audience.
The examples follow the Fable 5.2 launch, which emphasized improvements in coding and longer technical assignments. For creators, the practical opportunity is to test more ambitious ideas with less implementation work, while keeping time, token usage and repeated refinement visible alongside the result.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
