Cost calculator

Fable 5.1 vs GPT-6 Astra — per 1M tokens. Source: fable-5-1-vs-gpt-6-astra.md §1. Cache line is the real bill for agents (80-95% cache hit).

Presets (from doc)

Inputs

Input tokens

Output tokens

Cache hit % (of input)

Batch API (50% off) Input > 272K (Astra surcharge: 2× input/cache, 1.5× output) Include cache write (one-time prefix write at $12.50/M)

Rates per 1M: Input $10 / Output $50 / Cache read Astra $1.00 Fable $0.25 / Cache write $12.50. Fable has no over-threshold surcharge.

Result

GPT-6 Astra

Claude Fable 5.1

Cheaper

Astra Fable 5.1

How it is computed
cached = input × hit% → billed at cache-read rate; uncached = input - cached → billed at input rate; output billed at output rate; if batch, all ×0.5; if over-threshold, Astra input/cache ×2 and output ×1.5; optional cache-write = (cached / 1M) × 12.50 (× batch/surcharge as applicable).