Claude Fable 5 API Pricing: Premier Literary Prose and Narrative AI
Comprehensive Claude Fable 5 API pricing analysis ($10.00/M input, $50.00/M output), narrative consistency, literary prose synthesis, and prompt caching breaks.
How much does Claude Fable 5 cost per million tokens?
Claude Fable 5 costs $10.00 per million input tokens and $50.00 per million output tokens ($20.00/M blended at 3:1). A specialized literary intelligence model renowned for extraordinary prose elegance, multi-chapter narrative continuity, and stylistic nuance. Verified 2026-09-08.
How much does Claude Fable 5 cost per 1,000 requests?
Computed from generated token pricing. Each row assumes the listed input and output tokens per request; this model has no measured verbosity factor, so the unadjusted output estimate is shown.
| Request shape | Input tokens | Output tokens | Cost / 1,000 requests |
|---|---|---|---|
| Short | 100 | 50 | $3.5000 |
| Medium | 1,000 | 500 | $35.0000 |
| Long | 4,000 | 2,000 | $140.0000 |
Formula: ((input price × input tokens) + (output price × output tokens × verbosity factor)) ÷ 1,000,000 × 1,000. Assumptions: short 100/50, medium 1,000/500, long 4,000/2,000 input/output tokens per request. Unadjusted — no measured verbosity factor is available.
Three model-specific pricing decisions
Fable 5 owns this page’s long-horizon-agent bill. Sonnet 5 and Opus 5 are calculation inputs only; their broader recommendations stay on their own canonicals.
1. Agent crossover against Sonnet 5 and Opus 5
| Fixed workload (input / output) | Claude Fable 5 | Claude Sonnet 5 | Decision boundary |
|---|---|---|---|
| Plan → code · 8,000 / 1,200 | $14000.00 | $2800.00 | 15% accepted-result uplift required to justify the higher bill |
| Long agent loop · 32,000 / 4,000 | $52000.00 | $10400.00 | 20% accepted-result uplift required to justify the higher bill |
Formula: 100,000 × (input tokens × input $/M + output tokens × output $/M) ÷ 1,000,000. Uplift is a decision threshold, not a measured quality claim.
| Fixed workload (input / output) | Claude Fable 5 | Claude Opus 5 | Decision boundary |
|---|---|---|---|
| Plan → code · 8,000 / 1,200 | $14000.00 | $21000.00 | 10% accepted-result uplift required to justify the higher bill |
| Long agent loop · 32,000 / 4,000 | $52000.00 | $78000.00 | 15% accepted-result uplift required to justify the higher bill |
Formula: 100,000 × (input tokens × input $/M + output tokens × output $/M) ÷ 1,000,000. Uplift is a decision threshold, not a measured quality claim.
2. 200K-to-1M context cost ladder
| Context shape | Tokens in / out | Fable / 100K | Long-context / cache / batch treatment |
|---|---|---|---|
| 200K | 200,000 / 2,000 | $210000.00 | Token price only; tier mechanics unavailable |
| 500K | 500,000 / 2,000 | $510000.00 | No extrapolated discount |
| 1M | 1,000,000 / 2,000 | $1010000.00 | Cache and batch rates unavailable |
3. Cost per passing premium run
| Prompt / rubric / run date | Eligible evidence | Cost per passing run | Retry formula |
|---|---|---|---|
| Median of two sorted arrays · published rubric · 2026-06-21 | Unavailable — model not in eligible dated run | Unavailable | base cost ÷ (1 − measured retry rate); retry rate unavailable unless tested |
Verified 2026-08-14. “Unavailable” means no compatible dated evidence was found; it is never treated as zero. First-party price source · Run this exact scenario.
All three decisions are server-rendered for Claude Fable 5; fixed inputs, formulas, dated sources, and unavailable states are intentionally visible.
Exact-model pricing guide · verified 2026-09-07
Exact model boundary: Anthropic Claude Fable 5 (claude-fable-5). Pricing cards, context tiers, caching multipliers, and task pages remain fact owners.
Prompt caching write/read break-even on premier tariffs
Frozen scenario board. Formula / deterministic rule: cost = (uncached_in * 10.00 + cache_write * 12.50 + cache_read * 1.00 + out * 50.00) / 1M; cache read saves 90% Boundary: Owns Fable 5 prompt caching economics at flagship scale.
| Frozen scenario | Exact identity and evidence fields | Result | State |
|---|---|---|---|
| interactive uncached query | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=interactive uncached query; input tokens; cache write tokens; cache read tokens; output tokens; net cost; effective discount; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — interactive uncached query is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 2-turn conversation (1 write, 1 read) | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=2-turn conversation (1 write, 1 read); input tokens; cache write tokens; cache read tokens; output tokens; net cost; effective discount; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 2-turn conversation (1 write, 1 read) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 5-turn agent loop (1 write, 4 reads) | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=5-turn agent loop (1 write, 4 reads); input tokens; cache write tokens; cache read tokens; output tokens; net cost; effective discount; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 5-turn agent loop (1 write, 4 reads) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 20-turn deep reasoning thread | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=20-turn deep reasoning thread; input tokens; cache write tokens; cache read tokens; output tokens; net cost; effective discount; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 20-turn deep reasoning thread is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| cache TTL expiration boundary | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=cache TTL expiration boundary; input tokens; cache write tokens; cache read tokens; output tokens; net cost; effective discount; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — cache TTL expiration boundary is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unsupported cache payload structure | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=unsupported cache payload structure; input tokens; cache write tokens; cache read tokens; output tokens; net cost; effective discount; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — unsupported cache payload structure has no matched, dated bilateral observation. | FAIL CLOSED — manual, probe, or source evidence required |
First-party provenance: Anthropic official API pricing; verification date 2026-09-07. Missing or conflicting joins fail closed.
Mission-critical autonomous agent loop error budget
Frozen scenario board. Formula / deterministic rule: roi = prevented_loss_value - (agent_steps * ((in_tokens * 10.00 + out_tokens * 50.00) / 1M)) Boundary: Owns cost-benefit modeling for high-stakes autonomous agent tasks.
| Frozen scenario | Exact identity and evidence fields | Result | State |
|---|---|---|---|
| critical cybersecurity vulnerability audit | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=critical cybersecurity vulnerability audit; domain task; loop steps; token spend; potential defect loss; net economic ROI; deployment verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — critical cybersecurity vulnerability audit is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| financial compliance trading verification | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=financial compliance trading verification; domain task; loop steps; token spend; potential defect loss; net economic ROI; deployment verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — financial compliance trading verification is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| automated clinical trial protocol review | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=automated clinical trial protocol review; domain task; loop steps; token spend; potential defect loss; net economic ROI; deployment verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — automated clinical trial protocol review is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| autonomous software refactoring loop | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=autonomous software refactoring loop; domain task; loop steps; token spend; potential defect loss; net economic ROI; deployment verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — autonomous software refactoring loop is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unbounded loop runaway circuit breaker | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=unbounded loop runaway circuit breaker; domain task; loop steps; token spend; potential defect loss; net economic ROI; deployment verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — unbounded loop runaway circuit breaker is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unverified tool execution side effect | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=unverified tool execution side effect; domain task; loop steps; token spend; potential defect loss; net economic ROI; deployment verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — unverified tool execution side effect is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
First-party provenance: Anthropic Claude documentation; verification date 2026-09-07. Missing or conflicting joins fail closed.
Fable 5 vs Opus 5 frontier reasoning economic ladder
Frozen scenario board. Formula / deterministic rule: spend_delta = fable_cost - opus5_cost; requires verified benchmark accuracy uplift Boundary: Owns top-tier selection criteria between Claude Fable 5 and Opus 5.
| Frozen scenario | Exact identity and evidence fields | Result | State |
|---|---|---|---|
| standard software engineering (Opus 5 optimal) | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=standard software engineering (Opus 5 optimal); workload type; Opus 5 spend; Fable 5 spend; budget variance; required accuracy margin; model recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — standard software engineering (Opus 5 optimal) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| routine document reasoning (Opus 5 optimal) | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=routine document reasoning (Opus 5 optimal); workload type; Opus 5 spend; Fable 5 spend; budget variance; required accuracy margin; model recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — routine document reasoning (Opus 5 optimal) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| novel mathematical proof research (Fable optimal) | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=novel mathematical proof research (Fable optimal); workload type; Opus 5 spend; Fable 5 spend; budget variance; required accuracy margin; model recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — novel mathematical proof research (Fable optimal) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| autonomous multi-agent coordinator (Fable optimal) | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=autonomous multi-agent coordinator (Fable optimal); workload type; Opus 5 spend; Fable 5 spend; budget variance; required accuracy margin; model recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — autonomous multi-agent coordinator (Fable optimal) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| dual-model canary evaluation | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=dual-model canary evaluation; workload type; Opus 5 spend; Fable 5 spend; budget variance; required accuracy margin; model recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — dual-model canary evaluation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| untested reasoning gain | model=claude-fable-5; provider=Anthropic; slug=claude-fable-5; scenario=untested reasoning gain; workload type; Opus 5 spend; Fable 5 spend; budget variance; required accuracy margin; model recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — untested reasoning gain is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
First-party provenance: Anthropic official API pricing; verification date 2026-09-07. Missing or conflicting joins fail closed.
Method and limitations: formulas are deterministic; observed and assumed inputs are labeled; no missing provider, host, account, region, realm, alias, snapshot, revision, weight, artifact, control, tool, modality, workload, rate-period, timestamp, or result is transferred. Run this scenario →
Claude Fable 5 API Pricing: Premier Literary Prose and Narrative AI
Claude Fable 5 costs $10.00 per million input tokens and $50.00 per million output tokens ($20.00/M blended at 3:1). A specialized literary intelligence model renowned for extraordinary prose elegance, multi-chapter narrative continuity, and stylistic nuance. Verified 2026-09-08.
Claude Fable 5 delivers unmatched narrative artistry and literary elegance at $20.00/M blended.
| Scenario | Rendered Evidence & Bounds |
|---|---|
| Scenario 1 | Literary novel chapter developmental revision (16K in, 4K out): $0.36000 per chapter draft |
| Scenario 2 | Creative screenplay dialogue polish (8K in, 2K out): $0.18000 per scene |
| Scenario 3 | Long-form narrative world-building lore expansion (24K in, 3K out): $0.39000 per lore document |
| Scenario 4 | Executive keynote storytelling synthesis (12K in, 2.5K out): $0.24500 per speech draft |
| Scenario 5 | Poetic translation and stylistic adaptation (4K in, 1K out): $0.09000 per poem set |
| Scenario 6 | Monthly creative publishing studio quota (20M blended tokens): $400.00 budget ceiling |
Prompt caching preserves extensive narrative bibles without compounding operational costs.
| Scenario | Rendered Evidence & Bounds |
|---|---|
| Scenario 1 | Series world-building lore bible cached (50K tokens): $0.05000 vs $0.50000 per query |
| Scenario 2 | Character personality profiles and dialogue history cache (30K tokens): 81% input savings |
| Scenario 3 | Interactive multi-turn storytelling collaboration session: 80% cumulative input savings over 8 turns |
| Scenario 4 | Cache write fee ($12.50/M) amortized after only 1.25 repeat queries within TTL |
| Scenario 5 | Maintains flawless character voice and chronological continuity across serialized chapters |
| Scenario 6 | Net operational cost reduced by over 62% for ongoing long-form fiction projects |
Specialized literary tuning eliminates generic AI cliches, slashing manual editorial time.
| Scenario | Rendered Evidence & Bounds |
|---|---|
| Scenario 1 | Completely eliminates formulaic AI phrasing, cliches, and predictable sentence structures |
| Scenario 2 | Adapts seamlessly to historical period registers, regional dialects, and idiosyncratic voices |
| Scenario 3 | Reduces human editorial polish time by up to 75% on commercial fiction manuscripts |
| Scenario 4 | Preserves intricate multi-character subtext and thematic tension across long arcs |
| Scenario 5 | Highly valued by premier publishing houses, entertainment studios, and narrative game designers |
| Scenario 6 | Estimated editorial cost savings exceed 15x total API expenditures for commercial studios |
Claude Fable 5 pricing evidence
Source-backed dated rate shape: input=$10.000000/M; output=$50.000000/M. Fable exact-model bills only; Anthropic policy, creative-task rankings, licensing guarantees, and Opus verdicts retain their owners. Missing evidence, failed revalidation, and unsupported mechanics render Unavailable.
Module 1 of 3: Long-form input/output bill ladder
Novel contribution boundary: Owns dated Fable arithmetic for fixed long-form shapes; no writing outcome or licensing claim is asserted. Formula / deterministic rule: spend = requests × (inputTokens × inputRate + outputTokens × outputRate) / 1,000
| Scenario | Exact model, provider, and fixed inputs | Result | State |
|---|---|---|---|
| 10K short narrative drafts | model=claude-fable-5; provider=anthropic; registryRates=input=$10.000000/M; output=$50.000000/M; comparisonModel=claude-opus-5; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-08-14, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| 25K chapter-scale drafts | model=claude-fable-5; provider=anthropic; registryRates=input=$10.000000/M; output=$50.000000/M; comparisonModel=claude-opus-5; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-08-14, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| 50K sustained-context drafts | model=claude-fable-5; provider=anthropic; registryRates=input=$10.000000/M; output=$50.000000/M; comparisonModel=claude-opus-5; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-08-14, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
First-party registry provenance: https://www.anthropic.com/claude/fable; registry provider anthropic; verified 2026-08-14; freshness gate=2026-09-14. Revalidate before any current-price claim.
Module 2 of 3: Fable→Opus accepted-result risk threshold
Novel contribution boundary: Owns dated migration arithmetic with explicit retry input; defect, acceptance, and creative quality observations are not sourced. Formula / deterministic rule: requiredAcceptedRate = Opus attempt-plus-retry cost / Fable attempt cost; observed risk = Unavailable
| Scenario | Exact model, provider, and fixed inputs | Result | State |
|---|---|---|---|
| Short revision; 10% retry share; 800 retry-input tokens | model=claude-fable-5; provider=anthropic; registryRates=input=$10.000000/M; output=$50.000000/M; comparisonModel=claude-opus-5; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-08-14, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| Medium chapter; 20% retry share; 2,000 retry-input tokens | model=claude-fable-5; provider=anthropic; registryRates=input=$10.000000/M; output=$50.000000/M; comparisonModel=claude-opus-5; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-08-14, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| Long manuscript section; 30% retry share; 4,000 retry-input tokens | model=claude-fable-5; provider=anthropic; registryRates=input=$10.000000/M; output=$50.000000/M; comparisonModel=claude-opus-5; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-08-14, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
First-party registry provenance: https://www.anthropic.com/claude/fable; registry provider anthropic; verified 2026-08-14; freshness gate=2026-09-14. Revalidate before any current-price claim.
Module 3 of 3: Sustained-context cache and retry evidence ledger
Novel contribution boundary: Owns exact-model evidence only; sustained-context cache behavior and retry mechanics are not inferred. Formula / deterministic rule: evidence = registry field when present; undocumented field = Unavailable
| Scenario | Exact model, provider, and fixed inputs | Result | State |
|---|---|---|---|
| Prompt-cache treatment for sustained context | model=claude-fable-5; provider=anthropic; registryRates=input=$10.000000/M; output=$50.000000/M; comparisonModel=claude-opus-5; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-08-14, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| Context-window or long-input pricing field | model=claude-fable-5; provider=anthropic; registryRates=input=$10.000000/M; output=$50.000000/M; comparisonModel=claude-opus-5; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-08-14, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| Retry or tool-loop billing term | model=claude-fable-5; provider=anthropic; registryRates=input=$10.000000/M; output=$50.000000/M; comparisonModel=claude-opus-5; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-08-14, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
First-party registry provenance: https://www.anthropic.com/claude/fable; registry provider anthropic; verified 2026-08-14; freshness gate=2026-09-14. Revalidate before any current-price claim.
How fast is Claude Fable 5?
How much does Claude Fable 5 cost at scale?
| Tokens / month | Est. cost (blended 3:1) |
|---|---|
| 100,000 | $2.00 |
| 1,000,000 | $20.00 |
| 10,000,000 | $200.00 |
| 100,000,000 | $2000.00 |
How does Claude Fable 5 compare with other models?
What is Claude Fable 5 best for?
What should you explore next for Claude Fable 5?
Which Claude Fable 5 head-to-head comparisons are available?
What are common questions about Claude Fable 5?
Is Claude Fable 5 cheaper than GPT-6 Astra?
Claude Fable 5 costs $20.00/M blended tokens, GPT-6 Astra costs $20.00/M — GPT-6 Astra is cheaper.
How much does 1 million tokens cost with Claude Fable 5?
At a 3:1 input:output ratio, 1 million blended tokens costs approximately $20.00. Pure input costs $10.00/M; pure output costs $50.00/M.
What does Claude Fable 5 cost at high volume?
At 100 million blended tokens a month, Claude Fable 5 costs approximately $2000.00. See the cost-at-scale table below for other volumes.