Anthropic
Claude API
Anthropic's Claude line — three models split by capability and by writing style.
3
シリーズ内のモデル数
The Claude models we serve split along two axes rather than one: Sonnet and Opus are the balanced and the most-capable tiers, while Fable is tuned specifically for long-form writing and narrative work. All three speak the OpenAI protocol here and all support prompt caching, with cached reads at a tenth of fresh input.
概要
- 最低料金
- $1.01 / 100万トークン
このファミリーのモデル
| モデル | 最低料金 | モード | 開く |
|---|---|---|---|
| Claude Sonnet 5Anthropic's balanced model for coding, agentic workflows and everyday reasoning. Text and image input, long context. | $1.01/ 100万トークン | — | 開くドキュメント |
| Claude Opus 5Anthropic's most capable model for complex reasoning, coding and agentic work. | $3.02/ 100万トークン | — | 開くドキュメント |
| Claude Fable 5Anthropic model tuned for long-form writing and narrative work. | $3.63/ 100万トークン | — | 開くドキュメント |
料金
| モデル | オプション | 料金 |
|---|---|---|
| Claude Sonnet 5 | input | $1.01/ 100万トークン |
| Claude Sonnet 5 | output | $5.04/ 100万トークン |
| Claude Sonnet 5 | cacheRead | $0.101/ 100万トークン |
| Claude Sonnet 5 | cacheWrite | $1.18/ 100万トークン |
| Claude Opus 5 | input | $3.02/ 100万トークン |
| Claude Opus 5 | output | $15.12/ 100万トークン |
| Claude Opus 5 | cacheRead | $0.302/ 100万トークン |
| Claude Opus 5 | cacheWrite | $3.78/ 100万トークン |
| Claude Fable 5 | input | $3.63/ 100万トークン |
| Claude Fable 5 | output | $18.17/ 100万トークン |
| Claude Fable 5 | cacheRead | $0.363/ 100万トークン |
| Claude Fable 5 | cacheWrite | $4.54/ 100万トークン |
どれを選ぶか
Claude Sonnet 5
The default. Coding, agentic workflows and everyday reasoning with text and image input, at a third of Opus's price.
Claude Opus 5
Complex reasoning, hard coding problems and long agentic runs where capability matters more than cost.
Claude Fable 5
Long-form writing and narrative work — the tier tuned for prose rather than for reasoning benchmarks.
強み
- A writing-specialised tier alongside the general ones, which most families do not offer.
- Prompt caching on all three, with cached reads at a tenth of the input price.
- Sonnet takes image input as well as text, at the lowest price in the line.
- OpenAI-protocol compatible, so existing client code works unchanged.
制限
- Output tokens cost five times input across the line, so long generations are expensive.
- Cache writes are billed above fresh input, so caching only pays back on reuse.
- Fable is the most expensive tier per input token despite not being the most capable one.
提供元によるこのシリーズの説明: Anthropic — Claude