Superpower Daily
Resect AI Raises $25M to Try Changing Enterprise Models Before...Artificial Analysis Clarifies Astra’s 67-Point Coding Score Is...Nvidia Agrees to Buy Hugging Face for $12.93B, Promising Hardwa...Google Opens Lyria 3.5 to Developers and Adds Song Generation t...Arizona Study Finds Repeated Falsehoods Can Sway AI Models in L...NHTSA Opens Audit of About 1,000 Tesla Cybercabs as Austin Serv...Resect AI Raises $25M to Try Changing Enterprise Models Before...Artificial Analysis Clarifies Astra’s 67-Point Coding Score Is...Nvidia Agrees to Buy Hugging Face for $12.93B, Promising Hardwa...Google Opens Lyria 3.5 to Developers and Adds Song Generation t...Arizona Study Finds Repeated Falsehoods Can Sway AI Models in L...NHTSA Opens Audit of About 1,000 Tesla Cybercabs as Austin Serv...

Research investigation R0903 / comparison

Cache Economics Beyond the Headline: When Does Fable or Gemini Win on Repeated Context?

The lower cache-read price does not alone determine the lower-cost workflow: cache-write charges, eligibility, hit behavior, expiry, and output tokens can change the result.

Archived snapshotv2Sep 4, 2026
View latest dataset Read the original analysis
Verified observations
24

23 measured fields

Supported claims
9

9 material findings

Cited sources
11

11 primary or authoritative

Research score
87

Automated topic and evidence score

Interactive figureCache Economics Beyond the Headline: When Does...
CSV JSON
Data status1.875 verified records across 1 period

Snapshot only. There is not enough history to claim a trend yet.

Verified observationHover or focus any mark for exact valuesLast updated Sep 4, 2026

Version ledger

Frozen public editions

Each edition preserves the records, method, sources, and downloads available at publication time.

  1. v2 / latestSep 4, 202624 records / 11 sources

    Source-backed fields changed; the frozen snapshot contains 24 records.

  2. v1Sep 3, 202624 records / 11 sources

    Initial public snapshot with 24 records and 11 cited sources.

Coverage note

This synthesis uses primary provider documentation and derived calculations from the supplied rates. Artifact mode is evidence_matrix, so no chart is included.

Dataset ID
spd:cache-economics-beyond-the-headline-when-does-fable-or-gemini-win-on-repeated-context-b85ac7d5
Stable URL
/research/cache-economics-beyond-the-headline-when-does-fable-or-gemini-win-on-repeated-context-b85ac7d5
Version
v2
Coverage
2026-09-03
Records
24
Fields
7
Updated

Read the data

The records behind the figure

CSV JSON
Cache Economics Beyond the Headline: When Does Fable or Gemini Win on Repeated Context? data records
EntityMetricValueUnitObservedSourceTransform
Claude Fable 5.1100,000-token prefix cost, five-minute write plus 25 hits$1.875USD2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overview0.1*(12.5+0.25*25).
Gemini 3.8 Flash100,000-token prefix cost, initial request plus 25 implicit hits$0.2625USD2026-09-03https://ai.google.dev/gemini-api/docs/pricing0.1*(0.75+0.075*25).
Gemini 3.8 Flash100,000-token prefix cost, initial request plus 25 implicit misses$1.95USD2026-09-03https://ai.google.dev/gemini-api/docs/pricing0.1*(0.75+0.75*25).
Gemini 3.8 Flashcached-context price through 2026-12-31$0.075 per million tokensUSD/MTok2026-09-03https://ai.google.dev/gemini-api/docs/pricing—
Claude Fable 5.1cache-read price$0.25 per million tokensUSD/MTok2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overview—
Claude Fable 5.1five-minute cache-write price$12.50 per million tokensUSD/MTok2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overview—
Claude Fable 5.1 versus Gemini 3.8 FlashGemini hit fraction at which Fable loses for every reuse count74.074%percent of reuse tokens2026-09-03https://ai.google.dev/gemini-api/docs/pricingSolved 0.75*(1-h)+0.075*h=0.25; h=0.5/0.675.
Claude Fable 5.1minimum cacheable prefix512 tokenstokens2026-09-03https://platform.claude.com/docs/en/build-with-claude/prompt-caching—
Gemini 3.8 Flashminimum implicit-cache prefix4,096 tokenstokens2026-09-03https://ai.google.dev/gemini-api/docs/caching—
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, five-minute Fable cache, 0% Gemini hits24 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(11.75/(0.5-0.675*0))+1.
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, five-minute Fable cache, 25% Gemini hits36 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(11.75/(0.5-0.675*0.25))+1.
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, five-minute Fable cache, 50% Gemini hits73 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(11.75/(0.5-0.675*0.5))+1.
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, five-minute Fable cache, 60% Gemini hits124 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(11.75/(0.5-0.675*0.6))+1.
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, five-minute Fable cache, 70% Gemini hits428 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(11.75/(0.5-0.675*0.7))+1.
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, one-hour Fable cache, 0% Gemini hits39 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(19.25/(0.5-0.675*0))+1.
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, one-hour Fable cache, 25% Gemini hits59 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(19.25/(0.5-0.675*0.25))+1.
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, one-hour Fable cache, 50% Gemini hits119 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(19.25/(0.5-0.675*0.5))+1.
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, one-hour Fable cache, 60% Gemini hits203 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(19.25/(0.5-0.675*0.6))+1.
Claude Fable 5.1 versus Gemini 3.8 Flashminimum reuses, one-hour Fable cache, 70% Gemini hits701 reusesreuse requests2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overviewfloor(19.25/(0.5-0.675*0.7))+1; strict inequality requires 701 rather than 700.
Claude Fable 5.1one-hour cache-write price$20 per million tokensUSD/MTok2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overview—
Claude Fable 5.1prefix-cost formula with expiry clustersC*(0.25*N+19.75*K)USD when C is millions of tokens2026-09-03https://platform.claude.com/docs/en/build-with-claude/prompt-cachingExpanded C*(20*K+0.25*(N-K)); N is total requests and K is cache-write clusters.
Claude Fable 5.1prefix-cost formula with expiry clustersC*(0.25*N+12.25*K)USD when C is millions of tokens2026-09-03https://platform.claude.com/docs/en/build-with-claude/prompt-cachingExpanded C*(12.5*K+0.25*(N-K)); N is total requests and K is cache-write clusters.
Gemini 3.8 Flashstandard input price through 2026-12-31$0.75 per million tokensUSD/MTok2026-09-03https://ai.google.dev/gemini-api/docs/pricing—
Claude Fable 5.1uncached input price$10 per million tokensUSD/MTok2026-09-03https://platform.claude.com/docs/en/models/fable-5-1/overview—

Measurement technique

How to read this report

  1. 01Evidence matrix plan: reconcile provider price documentation, cache eligibility and expiry rules, API usage fields, and organization-level billing records where available.
  2. 02Compare only repeated-prefix input costs under equal billed prefix-token counts; exclude output and changing suffix tokens from crossover calculations.
  3. 03Use reported provider token fields for invoice reconciliation because the same content can tokenize differently between providers.
  4. 04Model Gemini implicit-cache behavior as a hit fraction rather than a guaranteed TTL-based outcome.
  5. 05Treat Fable cache writes as expiry clusters: a new cluster is required after expiry or invalidation.
Next report / 01AI Model Economics Index All research reports