Claude Fable 5

Anthropic·5 个数据来源·更新于 2026-11-10

上下文窗口1M厂商官网
最大输出128K厂商官网
输入 / 百万 Token$10厂商官网
输出 / 百万 Token$50厂商官网

评测表现

只在同一来源、同一口径下比较。切换来源可查看不同测试结果。

4 个评测来源

当前配置:各基准设置见下表

采集于 2026-11-10查看原始页面

综合分数和同榜图表来自 LLM Stats 榜单(采集于 2026-11-10);下方逐项成绩来自模型页。榜单出处

LLM Stats · LLM Stats Score
52.4原榜第 13 名

同榜模型表现

LLM Stats Score · 同一来源与榜单口径,分数越高越好

6 个配置
所选配置其他配置
查看完整对比数据
模型与配置LLM Stats Score每任务成本
GPT-6.1 Sol53.1未公布
Claude Opus 553未公布
Muse Wonder Full52.8未公布
Claude Fable 5 · 当前模型52.4未公布
GPT-5.6 Sol52.4未公布
Muse Wonder Full52.2未公布

不同来源的分数不能直接相加或横向比较;评测任务成本不是 API Token 单价。

逐项基准成绩

20 项已收录成绩 · 保留原始单位和测试条件

AutomationBench厂商自报 · 越高越好
17.4%

Zapier AutomationBench private held-out set at max effort. End-to-end business workflows across simulated company APIs.

成绩出处
BioMysteryBench厂商自报 · 越高越好
46.1%

Hard subset: 46.1%. Human-solved subset: 83.9%. Blocking safeguards cause Fable 5 to perform closer to Opus 4.8 due to fallbacks. System card reports this as a Mythos 5 or combined launch-table score rather than a separate Fable-only production score; Fable may fall back to Opus 4.8 when safeguards trigger.

成绩出处
Blueprint-Bench 2厂商自报 · 越高越好
38.6%

Spatial reasoning evaluation.

成绩出处
DeepSWE 1.1来源汇总 · 越高越好
70%

DeepSWE v1.1 leaderboard, mini-swe-agent harness, max effort; Pass@1 70% ± 4%.

成绩出处
ExploitBench厂商自报 · 越高越好
78%

Cybersecurity capture rate (Cap%). Blocking safeguards cause Fable 5 to perform closer to Opus 4.8 due to fallbacks; score reflects Claude Mythos 5. System card reports this as a Mythos 5 or combined launch-table score rather than a separate Fable-only production score; Fable may fall back to Opus 4.8 when safeguards trigger.

成绩出处
Finance Agent v2来源汇总 · 越高越好
56.31%
成绩出处
FrontierCode厂商自报 · 越高越好
46.3%

FrontierCode Main subset at xhigh reasoning effort. Score 46.3%, pass rate 48.8%; Fable 5 ranks #1 and outperforms every other model even at medium effort.

成绩出处
FrontierCode 1.1来源汇总 · 越高越好
53.5%

FrontierCode 1.1 current leaderboard; mergeability score at xhigh effort: 53.5%.

成绩出处
FrontierSWE来源汇总 · 越高越好
90%

Claude Code

成绩出处
GDP.pdf厂商自报 · 越高越好
29.8%

Document and vision-heavy knowledge-work evaluation, no tools.

成绩出处
HealthBench Professional厂商自报 · 越高越好
66%

Health evaluation. Blocking safeguards cause Fable 5 to perform closer to Opus 4.8 due to fallbacks. System card reports this as a Mythos 5 or combined launch-table score rather than a separate Fable-only production score; Fable may fall back to Opus 4.8 when safeguards trigger.

成绩出处
Humanity's Last Exam厂商自报 · 越高越好
64.5%

Multidisciplinary reasoning. With tools: 64.5%. Without tools: 59.0%. Blocking safeguards cause Fable 5 to perform closer to Opus 4.8 on some questions due to fallbacks. System card reports this as a Mythos 5 or combined launch-table score rather than a separate Fable-only production score; Fable may fall back to Opus 4.8 when safeguards trigger.

成绩出处
Legal Agent Benchmark厂商自报 · 越高越好
13.3%

Legal Agent Benchmark, Harvey's held-out set.

成绩出处
LifeSci Diamond (2026-11)来源汇总 · 越高越好
3.6%

Strict pass rate on 55 regulatory-affairs tasks. Plain API via Vercel AI Gateway with a web_search tool (Exa), provider default settings, one answer per task, graded by GPT-6.1 Sol at medium reasoning effort (Arca, 4 Nov 2026 edition).

成绩出处
LiveBench来源汇总 · 越高越好
78.31%

2026-01-08, Thinking xHigh Effort

成绩出处
OSWorld-Verified厂商自报 · 越高越好
85%

Fable 5 production-safeguarded configuration. OSWorld score reflects the updated harness with zoom-tool bug fix and 128K max tokens per turn.

成绩出处
SWE-Bench Pro厂商自报 · 越高越好
80%

Fable 5 production-safeguarded configuration on SWE-Bench Pro; average over 5 trials with thinking blocks included. Mythos 5 scored 80.3%.

成绩出处
SWE-Bench Verified厂商自报 · 越高越好
95%

Fable 5 production-safeguarded configuration on the 500-problem SWE-Bench Verified subset; average over 5 trials with thinking blocks included. Mythos 5 scored 95.5%.

成绩出处
Terminal-Bench 2.1厂商自报 · 越高越好
84.3%

Fable 5 production-safeguarded configuration on the mini-SWE-agent harness, high effort, 89 tasks x 5 attempts. 20.9% of trials hit a safety refusal and fell back to Claude Opus 4.8 for the rest of the trajectory; Mythos 5 scored 88.0%.

成绩出处
Terminal-Bench 4.0来源汇总 · 越高越好
44.5%

Terminal-Bench 4.0 public leaderboard; resolution rate 44.5% ± 3.8%; Claude Code agent; max effort.

成绩出处

技术规格

只展示有可靠来源的资料;优先采用厂商官网与官方模型仓库。

发布日期
2026-06-09厂商官网
知识截止
2026-01厂商官网
上下文窗口
1M Token厂商官网
最大输入
1M TokenLLM Stats
最大输出
128K Token厂商官网
输入模态
文本、图像厂商官网
输出模态
文本厂商官网
推理模式
支持厂商官网
许可证
ProprietaryLLM Stats

价格与用量

美元 / 百万 Token。不同服务商或测试配置的报价分别列出。

模型 API 价格,单位为美元每百万 Token
来源 / 服务商输入输出缓存读取
厂商官网 · Claude Fable 5$10$50$1
Anthropic$10$50$1
DeepInfra$10$50待补充

估算一次用量

按表格第一条报价计算,普通文本、无缓存。

预计费用$15
不含工具、图片、推理额外计费或阶梯价格;以服务商实际账单为准。

每个数字,都有出处

同名模型通过开发机构和明确版本关联。预览版、日期版本和不同规模模型分别建档。

厂商官网

厂商官方文档 · 基础规格与标准 API 报价

1 条记录最近采集 2026-11-10

LLM Stats

公开模型页与公开榜单;自报成绩单独标识

2 条记录最近采集 2026-11-10

BenchLM

公开评测数据;保留来源原始口径

1 条记录最近采集 2026-11-10

Arena

Arena 官方公开数据集 · CC BY 4.0

5 条记录最近采集 2026-11-10

OpenRouter

公开模型目录与 API 报价

1 条记录最近采集 2026-11-10
查看全部配置与采集记录
来源 / 配置采集时间原始数据
LLM StatsClaude Fable 52026-11-10T14:08:28Z查看来源
LLM StatsClaude Fable 5SHA-256 f91d6951d69f01a6…2026-11-10T13:53:13Z查看来源
BenchLMClaude Fable 52026-11-10查看来源
ArenaClaude Fable 5 (High)2026-11-10T14:07:52Z查看来源
ArenaClaude Fable 5 (High)2026-11-10T14:07:52Z查看来源
ArenaClaude Fable 5 (High)2026-11-10T14:07:52Z查看来源
ArenaClaude Fable 5 (High)2026-11-10T14:07:52Z查看来源
ArenaClaude Fable 5 (High)2026-11-10T14:07:52Z查看来源
OpenRouterClaude Fable 5SHA-256 63cbc45064d84060…2026-11-10T13:51:49Z查看来源
厂商官网Claude Fable 5SHA-256 2c1b733527d9a846…2026-11-10T14:06:46Z查看来源
继续浏览模型排行榜