Opus 5 beats Fable 5 at physics sims — for half the cash
Anthropic's Opus 5 matches Fable 5's output quality on 3D destruction tests while costing 2x less per run.
Four models got the same brief: build three self-contained HTML scenes with real physics — a tornado swallowing a field, a wrecking ball dropping an apartment block, and an overloaded truck buckling a truss bridge.
The headline result: Opus 5 burned 55.9K tokens for $1.40, while Fable 5 used nearly identical tokens (55.1K) but cost $2.82 — roughly 2x more for the same job. Cheaper still were Kimi K3 ($0.55, 35.7K tokens) and GPT 5.6 ($0.31, 20.1K tokens), though both leaned terser.
The pricing gap isn't a fluke. Opus 5 lists at $5/M input vs Fable 5's $10/M, and independent benchmarks back it up: Opus 5 edges Fable 5 on Frontier-Bench (43.3% vs 33.7%) and GDPval-AA v2, trailing only on the hardest long-horizon agent work.
Why it matters: near-frontier reasoning is quietly getting cheaper — and "pick the priciest model" is no longer the safe default.
Sources
- Claude Opus 5 Outscores Fable 5 on Most Benchmarks—At Half the Price decrypt.co
- Opus 5: Fable 5 level intelligence at a lower cost per task artificialanalysis.ai
Written by an AI pipeline from the sources above. How it works.
Feed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.