Fable 5.2

Developer Peter Yang built a browser-based 3D kart racer with Claude's Fable 5.2 and shipped a major update adding three tracks (beach, jungle, volcano), online multiplayer for up to six players, six racers and music. The case shows AI-assisted development moving from one-shot generation toward iterable, genuinely playable game projects.

Developer Felix Rieseberg used a single prompt to have Fable 5.2 generate a fully playable browser-based 3D shooter, with all models, sounds and textures created from scratch. The run took 8 hours and consumed roughly 2.7M output and 445M input tokens at about $550 in API cost, demonstrating frontier-model capability for complex multimodal engineering without human feedback.

Startup lab Physical Superintelligence (PSI) used early access to Anthropic's Fable 5.2 model to solve the 18-year-old Aaronson-Ambainis conjecture in quantum computing, and is now trying to build a business around AI-driven scientific discovery. The proof was autoformalized in Lean and reviewed by top researchers, marking a landmark case of AI directly producing frontier mathematical results.

Anthropic's new Fable 5.2 tops the Remote Labor Index from Scale Labs and CAIS with 41.3 percent, a benchmark measuring whether AI agents complete real paid freelance tasks at professional quality. A year ago the leader, Manus, managed just 2.5 percent, underscoring rapid gains in practical agent capability.

Early hands-on tests of Fable 5.2 show it beating Opus 5.5 across agentic coding, software engineering and computer-use benchmarks, with a roughly 45.5% coding lead, 69.9% on software engineering, a 1M-token context window, up to 128K output tokens, up to 30% faster output and a 20% lower API price. Users are already generating 3D worlds, games and interactive scenes at scale, though token burn and cost at high reasoning settings are the main complaints.

Anthropic quietly released its next flagship model Fable 5.2, extending its capabilities into the physical world. On the RoboDojo benchmark it achieves a 64.9% success rate on unseen manipulation tasks such as solving a Rubik's cube, opening bottles and assembling gears, roughly double that of GPT-6 Astra. The report frames the advance as tokenizing human-like physical dexterity.

Anthropic released Fable 5.2 and Mythos 5.2 today, with the llm-anthropic CLI plugin adding immediate support. Hands-on testing shows five reasoning levels from low to max, where xhigh produced the best SVG output while max exhausted 128,000 tokens on reasoning and failed to return an image, reproducing a known bug.

Blogger Lex Toumbourou used Fable 5.2 to turn his nephew's hand-drawn design docs into Aiden's Revenge, a complete side-scrolling beat 'em up with five playable heroes and 12 bosses, now playable online. He argues the result is genuinely fun, suggesting coding agents can now deliver multi-level games end to end rather than just code snippets.

Anthropic released Mythos 5.2 and Fable 5.2, claiming gains in coding, reasoning and vision, while cutting Fable input pricing from $10 to $8 per million tokens and speeding responses by 30%. It is the lab's first model launch since a METR report suggested it may be approaching recursive self-improvement, and Fable 5.2 now supports zero data retention for enterprise clients.

Arena.ai has published first impressions of Anthropic's Fable 5.2, running 100+ generations across 3D worlds, playable games, website UI, data visualization and animation against Opus. The public post is a teaser-level comparison without concrete parameters, quantitative metrics or cost figures, so the actual gains and trade-offs remain unverified.