Claude Fable 5
超大工程架构设计与深度重构的终极底牌；始终开启深度思考且延迟较高，适合异步复杂攻坚。
Score
62
Blended
¥144.0
per 1M tokens
Context
1M
Speed
70
tokens/sec
Latency
95.42s
Capability scores
- Overall
- 62
The overall figure is the Artificial Analysis Intelligence Index - a composite index, not a percentage. The current leader sits around 63. Figures are taken from the "with fallback" reasoning effort; the same model scores and responds differently at other efforts.
Pricing
| Item | Official price |
|---|---|
| Input | $10.0 |
| Output | $50.0 |
| Cache read | $1.00 |
| Cache write | $12.5 |
| Max output | 128K |
Per 1M tokens. Blended price works out to CNY 144.0 (USD converted at 1 USD = 7.2 CNY, rate checked 2026-08-27). A dash means the vendor does not publish that item or does not bill that way.
Good for
Strengths
- 综合分 62,仅次于 Opus 5
- 100 万 token 上下文
- 缓存命中价降到输入价的十分之一
Watch out for
- Claude 4.7 及之后的模型换用了新分词器,同样一段文本切出的 token 比 Sonnet 4.6 及更早版本多约 30%。因此按「每百万 token 单价」和其他厂商横向比价会低估实际花费。
- 缓存写入按有效期分档:5 分钟档 $12.5/百万 token,1 小时档 $20/百万 token。
- 首字延迟在榜单上高达 95 秒(始终开启思考),不适合要求即时响应的交互场景。
相关编程套餐
查看全量对比→每月 500 次 Fast Requests，超额后无限次慢速排队；支持多文件 Composer
每 5 小时约 45 条消息（根据全网负载动态浮动）；支持开启 Thinking 深度思考
充多少用多少，100% 按真实 Token 计费，无月费，无过期时间限制
Discussion & Comments
Share your thoughts, experiences and feedback.