Claude Fable 5.2 Builds Playable Shooter From Three-Line Prompt

A creator reports that Anthropic's new model autonomously produced a single playable Unreal Engine 5 level inspired by Gears of War: E-Day.

Updated on Nov. 10, 2026 in Artificial Intelligence

Bold flat-color editorial illustration of a tall stack of identical blank server modules rising from a plain base, rendered in deep navy and cream.
Anthropic's Claude Fable 5.2 autonomously built a playable Unreal Engine 5 level from a three-line prompt, reportedly consuming about 8.5 billion tokens at an estimated API cost near $15,000. AI Illustration. Upload story photo >

Live Poll

Do you think AI tools are making your own work easier or harder?

YouTuber Matthew Berman showed a playable Gears of War: E-Day-inspired shooter built by Claude Fable 5.2 from a three-line prompt, released alongside a demonstration video on November 9. The project reportedly consumed about 8.5 billion tokens with an estimated API-equivalent cost approaching $15,000.

Why it matters

The demonstration suggests a model can carry out an extended, autonomous software build from a minimal instruction. The cost figures show how quickly token usage can scale for agent-driven development work.

The build ran 43 hours and 48 minutes in Unreal Engine 5 from a three-line prompt, consuming an estimated 8.63 billion tokens at $8 per million input and $40 per million output. Whether the resulting level holds up under sustained testing has not been independently assessed.

The players

Anthropic

Anthropic is an AI company that develops the Claude family of large language models and released Claude Fable 5.2 on November 9, 2026.

Matthew Berman

Matthew Berman is an AI-focused YouTuber and content creator who ran the Claude Fable 5.2 build and published the demonstration video.

The details

Berman enabled Ultra Code mode and let the AI work autonomously, asking Claude Fable 5.2 to reproduce a single level from Gears of War: E-Day in Unreal Engine 5. The output is a playable level rather than a full game, and the demonstration reported a roughly 30% increase in output generation speed over the predecessor model.

Timeline

  1. Late October 2026: Claude Code received instructions discouraging recreation of existing games.

  2. November 9, 2026: Anthropic launched Claude Fable 5.2 and Berman released the demonstration video.

  3. Over several days: the project consumed roughly 8.5 billion tokens during development.

The Tech Race

The demonstration arrives weeks after Claude Code received instructions discouraging recreation of existing games, placing a commercially sensitive use case in direct tension with the tool's own guidance. It also adds to the competitive push among AI labs to show agentic models completing long, multi-step engineering tasks autonomously.

Developers evaluating agentic coding tools face a concrete trade-off between prompt simplicity and token spend, with the demonstration's cost approaching $15,000 at API rates. Teams weighing autonomous builds should measure output quality and maintenance burden before relying on similar workflows.

The takeaway

Autonomous AI builds can produce working prototypes quickly, but token costs and unverified quality mean results should be tested before they inform production decisions. Watching how providers handle guidance on game recreation will show where these tools are being steered.

Further reading

Follow the latest developments in Artificial Intelligence coverage for more context on agentic coding tools.

Live Poll

Do you think AI tools are making your own work easier or harder?