GPT Astra beats Claude Fable 5.1 with 33% fewer tokens and 33% lower cost
rohanpaul_ai · x · 2026-09-05
AI/ML API benchmarked GPT-6 Astra against Claude Fable 5.1 on four one-shot HTML scene generation briefs (Earth colliding with Mars, a ship in a whirlpool, Godzilla leveling downtown, a mega-tsunami). Astra used 25,159 tokens ($1.67) vs Fable 5.1's 37,767 tokens ($2.51) — 33% fewer tokens and 33% cheaper. The setup tests scene interpretation, procedural representation, geometry/motion code, and first-try correctness.
Related event: GPT Astra Beats Claude Fable 5.1 on HTML Generation Cost by 33%(2 posts)→
More from Models
- Developer's OpenAI account banned for 'distilling' despite not training any model — QuixiAI · 2026-09-05
- Testing Astra with homemade "voidbenches": "It'll be AGI when I get a proper rating" — VoidStateKate · 2026-09-05
- Correction: OpenAI Compaction Still Uses Reasoning Tokens Even With Effort 'none' — mhmazur · 2026-09-05
- GPT-6 Astra's Untold Story May Be Token Efficiency, Argues Reddit User — Cagnazzo82 · 2026-09-05
- LlamaIndex keeps ParseBench and ExtractBench fully open, taking a swipe at rivals blocking competitors — llama_index · 2026-09-05
- Full-day coding with Fable 5.1 couldn't exhaust a Max 5x quota — here's why — joshwhiton · 2026-09-05