GPT-6.2 Astra vs Claude Fable 5.2 on 3 real tasks: 13x cheaper and faster, but far less detail

Fresh-Resolution182 · reddit · 2026-11-20

The week GPT-6.2 Astra shipped, the author ran it head-to-head against Claude Fable 5.2 on three real creator tasks with identical prompts, assets, tokens and cost accounting: a three.js 3D platformer, an interactive jet-engine lesson page, and a 10-second 3D motion piece for a 41MB fox-girl GLB. No fix rounds—first output is what you see.

Results (both priced at 8 in / 40 out per million tokens):

| Task | Fable 5.2 | GPT-6.2 Astra |

|---|---|---|

| 3D platformer | 43,275 tokens, 428s, $1.74, collected 11/12 coins and reached the flag | 3,056 tokens, 31s, $0.13, collected 3, still on first platforms |

| Jet-engine lesson | 45,386 tokens, 469s, $1.82, added hover labels, highlight boxes, per-step data card | 3,457 tokens, 37s, $0.14, fuller cutaway but two cards have overlapping text |

| 3D motion piece | 40,171 tokens, 403s, $1.62, camera push-in, orbit, confetti, title card | 2,137 tokens, 23s, $0.09, clean studio orbit, quieter camera |

Since pricing is identical, the 13x cost and 14x time gap is purely how much Fable writes—and that extra code shows up as more detail. Astra is the fast cheap draft; Fable has more of what you'd want to ship.

Original post →

More from Models

Models channel →