Descriptions:
David Ondrej shares an intensive hands-on exploration of Fable 5, Anthropic’s newest and most capable model, which he describes as the largest quality leap in AI since the release of GPT-4. After testing the model for over 12 hours straight, he offers practical guidance that goes well beyond benchmark coverage.
A central focus is where and how to actually run Fable 5. Ondrej advises against using it through the Claude.ai app (too many guardrails) or raw API calls (no fallback when safety filters trigger), and recommends Cursor’s agent window or Claude Code as the two best environments. He details a specific prompting mindset shift required for the model — keeping prompts less prescriptive to let the model’s superior reasoning take over — and shares techniques to avoid the model’s tendency to downgrade itself to Opus 4.8 when safety thresholds are crossed.
On the capability side, Ondrej demonstrates building AutoGit, an open-source npm tool for automated git workflows, live during the video. He also shows Fable 5 generating 3D browser games inside Claude Artifacts, analyzing and ranking startup ideas, and revamping internal software systems. He notes the model is notably less agreeable than predecessors — more likely to push back on bad ideas and identify patterns the user hasn’t noticed. Pricing is flagged as double that of Opus, making cost management a practical concern for heavy users.
📺 Source: David Ondrej · Published June 10, 2026
🏷️ Format: Hands On Build
