Claude Fable 5.1
Anthropic's top-of-range model for demanding reasoning. On this index it no longer leads. Opus 5.5 and Sonnet 5.5 both passed it after it launched.
51Index · High
$3.91Per measured task
13.0 pts/$Points / $
51Tokens / sec
Every published effort
| Setting | Index | Per task | Tokens / sec |
|---|---|---|---|
| Max | 53 | $7.63 | 69 |
| Extra high | 53 | $5.98 | 61 |
| High | 51 | $3.91 | 51 |
| Medium | 49 | $2.98 | 50 |
| Low | 47 | $2.37 | 47 |
What this workload costs
List price for one call. Reasoning tokens count as output. A cache hit uses the lab’s cache rate when one is published.
Claude Fable 5.1
Cache appliedSpecs
- API id
- claude-fable-5-1
- Released
- 1 Sep 2026
- Context
- 1M
- Max output
- 128k
- Input / 1M
- $10.00
- Output / 1M
- $50.00
- Cache read / 1M
- $0.25
- Knowledge
- Jun 2026
- Inputs
- Text, Image
- Weights
- Closed
Cache reads are 2.5% of input. The sticker matches GPT-6 Astra. On a long cached agent run the bill is closer to Opus than the sticker suggests, and still higher.
Use it when
The hardest long-horizon work, when Anthropic's own guidance points you at Fable rather than Opus. Cache-heavy agent loops, where the $0.25 cache read matters.
Leave it when
As a default. At high effort, Opus scores higher on this index and costs less per measured task.