22,022 API calls measured: Claude Fable 5.1 uses 31% more tokens but costs 31% less per prompt
tenequm · reddit · 2026-09-03
The author analyzed 21 days / 22,022 API calls from their own Claude Code session archive (pond), comparing Fable 5 vs 5.1 per-prompt.
Key findings
- Fable 5.1 consumes 31% more tokens per prompt than 5, mostly cache reads
- But 5.1 bills cache reads at 25% of list price, so average per-prompt cost drops from $1.52 to $1.05 (31% cheaper)
Why the weekly limit fills faster? Three untestable theories:
- Plan limit math may not apply Fable 5.1's cache-read discounts
- The 5h window may have changed from 1/6 to 1/5 of the week
- Effort may affect 5.1 differently
Actionable tips
- Official prompting guide notes 5.1 tends to issue one tool call per turn; a nudge like "first privately list what you need, then request all non-dependent items in one response" encourages batching
- The author added this to their CLAUDE.md
Queries are available on request for others to reproduce the analysis on their own l transcripts.
More from coding & agent
- Burkov: if you don't interrogate Codex line-by-line at handoff, you get 80%-done work — burkov · 2026-10-21
- Mintlify launches Autopilot, an always-on agent that updates docs as you ship — stuffyokodraws · 2026-10-21
- Skills vs MCP vs RAG vs Memory: the four knowledge layers production agents need — MaryamMiradi · 2026-10-21
- When customer service bans AI agents, one goes incognito to fight Xfinity — MoonL88537 · 2026-10-21
- Every.to publishes raw interviews so readers can agent-check quotes in its Codex story — every · 2026-10-21
- One-prompt Blender: Voyager + Claude Opus 5.5 auto-connects to Blender and Unity — ravisparikh · 2026-10-21