Grok 4.5vsClaude Fable 5
在 14 个共同 benchmark 中，Claude Fable 5 整体领先：Grok 4.5 领先 1 项，Claude Fable 5 领先 13 项，持平 0 项，平均分差 -46.96。
Grok 4.5
xAI · 2026-07-08 · 编程大模型
Claude Fable 5
Anthropic · 2026-06-09 · 推理大模型
Grok 4.51 项(7%)(93%)13 项Claude Fable 5
评测分数
按能力类目分组，每组内按分差大小排列；共 14 项。
Coding and Software Engineer
Claude Fable 5 领先 5/5| 评测项 | Grok 4.5 | Claude Fable 5 | 分差 |
|---|---|---|---|
| DeepSWE | 5318 / 27Thinking High (With Tools) | 702 / 27Deep Thinking (With Tools) | -17 |
| SWE-Bench Pro - Public | 64.706 / 57Thinking High (With Tools) | 80.301 / 57Deep Thinking (With Tools) | -15.60 |
| FrontierCode 1.1 | 56.604 / 5Thinking High (With Tools) | 63.601 / 5Max (With Tools) | -7 |
| APEX-SWE | 53.603 / 3Thinking High (With Tools) | 58.801 / 3Max (With Tools) | -5.20 |
| CursorBench 3.2 | 66.704 / 4Thinking High (With Tools) | 70.501 / 4Max (With Tools) | -3.80 |
Productivity Knowledge
Claude Fable 5 领先 2/3| 评测项 | Grok 4.5 | Claude Fable 5 | 分差 |
|---|---|---|---|
| AA-Briefcase | 1,3136 / 6Thinking High (With Tools) | 1,5743 / 6Max (With Tools) | -261 |
| GDPval-AA v2 | 1,5267 / 13Thinking High (With Tools) | 1,7414 / 13Max (With Tools) | -215 |
| Harvey Lab-AA | 12.904 / 6Thinking High (With Tools) | 11.305 / 6Max (With Tools) | +1.60 |
AI Agent - Tool Usage
Claude Fable 5 领先 2/2| 评测项 | Grok 4.5 | Claude Fable 5 | 分差 |
|---|---|---|---|
| Terminal-Bench 3.0 | 15.705 / 6Thinking High (With Tools) | 34.102 / 6Max (With Tools) | -18.40 |
| Terminal-Bench 2.1 | 83.3014 / 44Thinking High (With Tools) | 884 / 44Deep Thinking (With Tools) | -4.70 |
Math and Reasoning
Claude Fable 5 领先 2/2| 评测项 | Grok 4.5 | Claude Fable 5 | 分差 |
|---|---|---|---|
| FrontierMath Tier 4 v2 | 24.3920 / 34Thinking High (No Tools) | 87.801 / 34最高（无工具） | -63.41 |
| FrontierMath v2 | 57.1921 / 34Thinking High (No Tools) | 87.023 / 34最高（无工具） | -29.82 |
Agent Level Benchmark
Claude Fable 5 领先 1/1| 评测项 | Grok 4.5 | Claude Fable 5 | 分差 |
|---|---|---|---|
| APEX-Agents | 47.104 / 5Thinking High (With Tools) | 59.201 / 5Max (With Tools) | -12.10 |
General Knowledge
Claude Fable 5 领先 1/1| 评测项 | Grok 4.5 | Claude Fable 5 | 分差 |
|---|---|---|---|
| AA Intelligence Index | 564 / 8Thinking High (With Tools) | 621 / 8Max (With Tools) | -6 |
规格对比
| 字段 | Grok 4.5 | Claude Fable 5 |
|---|---|---|
| 发布机构 | xAI | Anthropic |
| 发布时间 | 2026-07-08 | 2026-06-09 |
| 模型类型 | 编程大模型 | 推理大模型 |
| 架构 | 稠密模型 | 稠密模型 |
| 参数规模 | 暂无数据 | 暂无数据 |
| 上下文长度 | 500K | 1M |
| 最大输出 | 暂无数据 | 128K |
API 调用价格
价格优先使用 DataLearner 配置的 API 记录；缺失项不做推测。
| 价格项 | Grok 4.5 | Claude Fable 5 |
|---|---|---|
| 文本输入 | $2 / 1M tokens | $10 / 1M tokens |
| 文本输出 | $6 / 1M tokens | $50 / 1M tokens |
| 缓存读取 | $0.5 / 1M tokens | $1 / 1M tokens |
| 缓存写入 | 暂无公开价格 | $12.5 / 1M tokens |
小结
- Claude Fable 5在以下类目领先:Coding and Software Engineer (5/5)、Productivity Knowledge (2/3)、AI Agent - Tool Usage (2/2)、Math and Reasoning (2/2)、Agent Level Benchmark (1/1)、General Knowledge (1/1)
14 个共同 benchmark 上，Claude Fable 5 平均高出 46.96 分。
单项差距最大的 benchmark：AA-Briefcase — Grok 4.5 1,313，Claude Fable 5 1,574（分差 -261）。
本页正文由结构化模型、价格与 benchmark 数据生成，不使用实时 LLM 撰写。