Claude Fable 5.1 API pricing keeps $10.00/1M input and $50.00/1M output per 1M tokens. The meaningful change versus Fable 5 is cache reads at $0.250/1M (2.5% of input), not 10%. Rates in TokenCalculator checked 17 Sep 2026.
Price a Claude workload
In-browser Claude counts are Approx. Use the cost calculator for planning, then confirm critical budgets with Anthropic’s count_tokens API.
Claude Fable 5.1 rate card
API model id: claude-fable-5-1. Context window: 1M tokens at standard rates across the window. Max output: 128K. Anthropic publishes a separate 1-hour cache write tier ($20 / 1M) in official docs; TokenCalculator stores the 5-minute write line when one cache write field is available.
| Line | USD per 1M tokens |
|---|---|
| Input | $10.00/1M |
| Cached input (hits / refreshes) | $0.250/1M |
| Cache write (5m, catalog) | $12.50/1M |
| Output | $50.00/1M |
| Batch input | $5.00/1M |
| Batch output | $25.00/1M |
Why the cache read cut matters
Most Claude models price cache hits at 10% of input. Fable 5.1 and Mythos 5.1 use 2.5%. On agent loops that reread a large stable prefix, Anthropic estimates roughly 25% lower cost for typical work and up to about 45% for highly agentic sessions versus Fable 5. Your savings track cache hit share, not sticker input alone.
Fable 5.1 versus Opus 5 and Sonnet 5
| Model | Input / 1M | Output / 1M | Cache read / 1M |
|---|---|---|---|
| Claude Fable 5.1 | $10.00/1M | $50.00/1M | $0.250/1M |
| Claude Opus 5 | $5.00/1M | $25.00/1M | $0.500/1M |
| Claude Sonnet 5 | $2.00 | $10.00 | $0.20 |
Anthropic positions Opus 5 as the default frontier pick for most complex work. Use Fable 5.1 when evals still fall short on long-horizon agents or hard research, and you accept $10 / $50 list rates. Sonnet 5 remains the volume production default at $2 / $10.
Worked examples
- Fresh 10K input + 2K output about $0.20 on standard rates.
- Same turn with 8K of that input cached: cache read $0.002 + 2K fresh $0.02 + 2K output $0.10 about $0.122.
- Batch halves input and output when overnight latency is fine.
Tokenizer note
Anthropic notes that tokenizers from Opus 4.7 onward can produce roughly 30% more tokens on the same text than older Claude tokenizers. Budget from counted tokens, not Word counts, and treat TokenCalculator Approx labels honestly.
Common mistakes
- Assuming Fable 5.1 is cheaper than Astra on every workload because cache is cheaper (short uncached prompts tie at $10 / $50)
- Ignoring cache write cost on the first pass of a large prefix
- Using OpenAI Exact counts to budget Claude spend
- Picking Fable for routine chat that Sonnet 5 would finish cheaper
Frequently asked questions
How much does Claude Fable 5.1 cost?
Standard Claude API pricing is $10.00/1M per 1M input and $50.00/1M per 1M output. Cache reads are $0.250/1M.
What is the Claude Fable 5.1 model id?
Use claude-fable-5-1 on the Claude API.
Did Anthropic lower the sticker price for 5.1?
No. Input and output stay $10 / $50. The cut is on cache reads versus Fable 5.
When should I use Opus 5 instead?
Start with Opus 5 for most complex agentic coding. Move to Fable 5.1 when Opus still fails your evals at high effort or you need the longer-horizon Fable class.
Is there a long-context surcharge?
Anthropic documents standard per-token pricing across the 1M window for Fable 5.1. There is no Astra-style 272K cliff on the published Fable card.
Are browser token counts Exact for Claude?
No. TokenCalculator labels Claude Approx. Confirm shipping budgets with Anthropic count_tokens.
How does Fable 5.1 compare to GPT-6 Astra?
Same sticker. Fable wins on cache reads. Astra may differ on measured cost per task when reasoning token spend diverges. See the head-to-head guide.
Are these rates live?
No. Curated list prices with lastVerified 17 Sep 2026. Confirm on Anthropic’s pricing page.
Next steps
- GPT-6 Astra vs Claude Fable 5.1
- GPT-6 Astra pricing
- Claude API pricing hub
- Prompt caching guide
- Official Anthropic pricing
Related Sep 2026 pricing guides
Frontier rate cards, cost-per-task math, and long-context cliffs. Cross-link these when you compare Sol, Luna, Opus 5.5, Grok 4.7, or Astra.