Fable 5.1 burned 668K tokens answering one spec question, and there's no way to audit why
joshalbrecht · x · 2026-09-03
Developer joshalbrecht reports Fable 5.1 spent 668K tokens to answer a single question about a spec. He questions what the tokens were spent on and whether it was necessary — but the closed model offers no way to inspect its internal process, highlighting the opacity of closed-source agent cost accounting.
More from Models
- Meta Muse Spark 1.3 early coding tests impress: near-frontier quality at a third of the price — DeryaTR_ · 2026-09-03
- Muse Spark 1.3 debuts third on Stata Benchmark, beating everything but Claude Fable — armand_ruiz · 2026-09-03
- Meta releases Muse Spark 1.3: frontier performance "almost too cheap to meter" — denny_zhou · 2026-09-03
- GLM-OCR reads Nvidia's 61-page 10-Q at 2,086 tok/s for under 2 cents — spillai · 2026-09-03
- Alexandr Wang Touts Meta Muse Spark 1.3: 1 Minute vs Fable 5.1's 70 Minutes and $13 — alexandr_wang · 2026-09-03
- A Gemini Flash model reportedly tops the DeepSWE coding leaderboard — sunjiao123sun_ · 2026-09-03