Claude Fable 5.1 Should Not Be This Good (way better than Fable 5)

Summary In this video, creator Zo tests Anthropic’s newly released Claude Fable 5.1 by challenging the model to write code for three playable games from scratch without external game engines. Across single-file HTML implementations, Fable 5.1 builds a browser voxel engine modeled after Minecraft, a 2D lane-defense clone of Plants vs. Zombies, and a 3D procedural New York City Spider-Man web-swinging prototype using Three.js.

What is shown

• Benchmark overview [00:02]: Anthropic announcement table showing Claude Fable 5.1 benchmarks against Fable 5, Opus 5, and GPT-5.4 Sol (e.g., 52.6% on Terminal-Bench Science, 55.8% on Terminal-Bench 4.0, and 71.4% on SWE-bench 3.0).
• Minecraft generation and gameplay [01:03 - 04:26]: A master prompt asking Fable 5.1 (set to High effort) to create a self-contained Three.js Minecraft clone in one HTML file. After a ~42-minute autonomous coding run, Zo tests terrain generation, voxel mining, tool crafting at a crafting table, and creative mode house-building with custom procedural textures.
• Plants vs. Zombies recreation [04:58 - 08:55]: Zo submits a prompt for a complete lane-defense game titled Plants vs. Zombies: Backyard Siege without external image assets. The model generates 2D procedural sprites, sunflower economies, peashooters, wall-nuts, melon-pults, and multi-wave zombie battles culminating in a Brute boss fight and a "Lawn Defended" screen.
• 3D Spider-Man Web-Swinging [09:52 - 13:36]: Fable 5.1 is set to Ultracode/Max effort to build a 3D procedural NYC with pendulum rope physics and wall-running. After an initial clunky build, Zo inputs a second refinement prompt tuning anchor-point logic and swing velocity, resulting in high-speed swinging through procedural skyscrapers and views of the Brooklyn Bridge.

Claims & numbers

• The presenter notes Anthropic’s benchmark table rates Fable 5.1 at 52.6% on Terminal-Bench Science 0.1, 55.8% on Terminal-Bench 4.0 (with Mythos 5.1 reaching 60.9%), 1853 on GDPval-AA v2, 77.9% partial / 41.7% strict on OSWorld 2.0, and 71.4% on SWE-bench 3.0 [00:05 - 00:11].
• The presenter states the Minecraft coding run took approximately 42 minutes, 27.3k tokens, and consumed 44% of his 5-hour Claude limit on a 20x plan [01:16 - 01:21].
• The presenter rates the three generated games: Minecraft at 9.5/10 [12:28], Plants vs. Zombies at 8.5/10 [12:34], and Spider-Man Web-Swinging at 7/10 [12:45].

Notable quotes

• [00:35] "And trust me when I say, Fable 5.1 shocked me, especially on the last one."
• [08:33] "Like someone like me who cannot code at all, I just recreated one of my favorite games from childhood..."
• [12:56] "Making something that actually works is basically solved. But making something that feels right for the player... that's the real challenge with AI."

Assessment This is an authentic, independent hands-on community review and stress-test of Claude Fable 5.1's coding capabilities using Claude Code. While the generation process is sped up and edited down for pacing, the gameplay sessions and user interface prompts demonstrate genuine, working single-file code outputs produced by the model.

Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.