Claude Fable 5.1 Batch API Cuts Inference to $5/$25 Per Million Tokens
Anthropic has made Claude Fable 5.1 available through a batch API at $5 per million input tokens and $25 per million output tokens — half the real-time rate of $10/$50. The batch tier is already live on OpenRouter. The 98% cache discount that applies to standard Fable 5.1 calls remains in effect for batch requests.
What the Numbers Mean in Practice
At standard Fable 5.1 pricing, a cached read costs roughly $0.20 per million input tokens (98% off $10). Under batch pricing, that drops to $0.10/M. For pipelines that keep a large system prompt or document corpus in context — document review, code analysis across a large repo, multi-step reasoning chains with repeated context reloads — the effective per-token cost for non-novel input tokens falls below a tenth of a cent per thousand tokens.
Price comparison across the Fable family and batch tier:
| Model | Input /M | Output /M | Cache Discount |
|---|---|---|---|
| Claude Fable 5.1 (batch) | $5.00 | $25.00 | 98% |
| Claude Fable 5.1 (standard) | $10.00 | $50.00 | 98% |
| Claude Opus 5 | $15.00 | $75.00 | — |
Effective cached-context read cost in batch: $0.10/M input tokens.
The Artificial Analysis Intelligence Index rates Claude Fable 5.1 at 66, placing it first on the leaderboard and nearly double the median score (36) among reasoning models in a comparable price tier. This is the top-performing model on the Intelligence Index with the batch tier now available.
Trade-offs
Batch is asynchronous. Requests queue and may return within minutes or hours depending on load. It is the correct tier for scheduled inference jobs, overnight document processing, evaluation pipelines, and regulatory review workflows — not interactive features or anything requiring a real-time SLA.
The 1M-token context window is preserved in batch mode. For workloads that routinely push toward context limits, Fable 5.1 batch combines extreme-length context with a cost structure that competes with mid-tier models on price.
Positioning in the Anthropic Lineup
Claude Fable 5.1 launched September 1 already 25% cheaper than Fable 5 for standard token-billed workloads, with savings up to 45% in cache-heavy pipelines. The batch tier stacks on top of that reduction.
For async-tolerant workflows, the effective conclusion is that Fable 5.1 batch ends up cheaper than Claude Sonnet 5 at real-time pricing while sitting multiple benchmark points above it across Terminal-Bench-Science, GDPval-AA v2, OSWorld 2.0, and CursorBench. The intended positioning: pay mid-tier rates for frontier-tier capability, at the cost of latency.
1M Context at Batch Cost
One practical consequence not obvious from the headline: Fable 5.1 supports a 1M-token context window. Running a 500K-token context at batch pricing costs $2.50 per call in input tokens before caching, falling to $0.05 per repeated call after the initial cache warm. For document-intensive use cases — legal discovery, clinical record review, large codebase analysis — those numbers make Fable 5.1 batch competitive with purpose-built retrieval systems that add infrastructure complexity.
Batch API access is available now via OpenRouter and directly through Anthropic’s API.