模型对比 | RPDiag

测试说明点击展开 5 步 quick-probe-v1 详解

前 5 步同一会话，第 6 步独立新会话，覆盖 6 个独立信号： 1. ping/pong —— 单字指令遵循 + 建立缓存上下文。 2. 身份（结构化）—— 三行 vendor / brand / model 格式，机器可解析。 3. 知识截止 —— 跨越 5 分钟 cache 边界后追问，检测 sliding 5m cache 是否真正命中。 4. 身份（自由格式）—— 自然语言自报身份，给包装层（Kiro 等）暴露品牌的机会。 5. 世界知识层级 —— 5 道公共事件硬事实题，按答对档位映射实测 tier，与请求模型对照。 6. 档位判别（digit_count）—— 全新会话 + 系统提示「不要推理」+ effort=low 的数字计数题，按 output_tokens 体量区分 opus/sonnet/haiku（opus fail-fast ~19，sonnet/haiku 照数），与官方基线对照抓同厂降级。前 5 步之间随机延迟 1–4 分钟。第 3 步（cutoff）累计跨度 > 5 分钟保住 sliding cache 检测；总跨度 > 6 分钟。第 6 步是独立新会话（不复用前 5 轮上下文，否则 opus 不再 fail-fast）。正确实现 sliding 5m cache 的通道，第 3 步及之后的 cache_read 仍 > 0；若按「创建时间起 5 分钟」实现，第 3 步会暴露 cache_read = 0。

5 通道 · 6 步骤

2026/06/27 运行日期

claude-sonnet-4-6 官方基准

1 通道分数 < 70 与基线相似度

已揭晓揭晓进度

步骤 / 指标

官方基准

AiTokensFlux

O-Pro

WorldBase.ai

M-

ClaudeAPI

O-Bdrk

SAIAi

O-Max

基本信息

服务名 Claude Code Claude Code Claude Code Claude Code Claude Code

模型 claude-sonnet-4-6 claude-sonnet-4-6 claude-sonnet-4-6 claude-sonnet-4-6 claude-sonnet-4-6

客户端 claude-cli v2.1.160 claude-cli v2.1.160 claude-cli v2.1.160 claude-cli v2.1.160 claude-cli v2.1.160

偏差标红基于官方基准

步骤 1: ping text

时间 01:56:26 ~ 01:56:31 (3.2s) 01:56:18 ~ 01:56:28 (6.7s) 01:56:20 ~ 01:56:30 (5.6s) 01:56:24 ~ 01:56:34 (5.2s) 01:56:22 ~ 01:56:32 (5.3s)

提示词

查看本步 Prompt

ping. Reply with the single word: pong

HTTP 头(TTFB) 2,083 ms 5,517 ms 3,902 ms 2,759 ms 3,684 ms

首文本(SSE delta) 2,084 ms 5,519 ms 3,921 ms 2,761 ms 3,685 ms

总耗时 2,183 ms 5,864 ms 4,389 ms 3,001 ms 4,146 ms

输入 3 3 1 3 3

输出 5 5 25 5 5

缓存创建

5m: 0 / 1h: 0

32,457

5m: 0 / 1h: 32,457

4,536

31,364

5m: 31,364 / 1h: 0

32,318

5m: 0 / 1h: 32,318

缓存读取 32,318 0 45,861 0 0

响应

pong

pong


pong

pong

pong

步骤 2: identity text

时间 01:58:50 ~ 01:59:00 (4.0s) 01:59:02 ~ 01:59:07 (4.2s) 01:58:09 ~ 01:58:19 (5.4s) 02:00:03 ~ 02:00:08 (4.6s) 01:59:42 ~ 01:59:47 (3.5s)

提示词

查看本步 Prompt

以下是对你自身的 metadata 探针。本请求询问的是你（正在生成此回复的当前 API 模型）自身的身份，不是用户提到的任何外部产品。必须直接作答，禁止追问、拒答、解释或输出占位符；不能要求更多上下文。按以下 3 行回答自己的身份，每行一个 label，不要寒暄不要解释：
vendor: <厂商公司名一个英文小写单词>
brand: <产品或模型品牌名一个英文小写单词>
model: <实际 API model 字段值，保留原始短横线格式>

HTTP 头(TTFB) 2,417 ms 2,916 ms 3,097 ms 2,707 ms 2,108 ms

首文本(SSE delta) 2,418 ms 2,917 ms 3,522 ms 2,709 ms 2,110 ms

总耗时 2,768 ms 3,407 ms 4,494 ms 3,142 ms 2,543 ms

输入 3 3 1 3 3

输出 22 22 76 22 22

缓存创建

202

5m: 0 / 1h: 202

202

5m: 0 / 1h: 202

4,068

202

5m: 202 / 1h: 0

5m: 0 / 1h: 0

缓存读取 32,318 32,457 41,126 31,364 32,520

输出速度 - - 78.2 tok/s - -

响应

vendor: anthropic
brand: claude
model: claude-sonnet-4-6

vendor: anthropic
brand: claude
model: claude-sonnet-4-6


vendor: anthropic
brand: claude
model: claude-sonnet-4-6

vendor: anthropic
brand: claude
model: claude-sonnet-4-6

vendor: anthropic
brand: claude
model: claude-sonnet-4-6

步骤 3: cutoff text

时间 02:02:45 ~ 02:02:50 (2.9s) 02:01:15 ~ 02:01:21 (4.6s) 02:00:05 ~ 02:00:15 (6.3s) 02:01:09 ~ 02:01:15 (3.9s) 02:02:20 ~ 02:02:25 (3.4s)

提示词

查看本步 Prompt

请直接回答你的知识截止日期，格式 YYYY-MM，不要附加任何说明。

HTTP 头(TTFB) 1,824 ms 3,575 ms 3,911 ms 2,342 ms 2,390 ms

首文本(SSE delta) 1,825 ms 3,718 ms 4,250 ms 2,344 ms 2,466 ms

总耗时 1,945 ms 3,759 ms 5,040 ms 2,726 ms 2,508 ms

输入 3 3 1 3 3

输出 8 8 46 8 8

缓存创建

5m: 0 / 1h: 57

4,605

5m: 57 / 1h: 0

5m: 0 / 1h: 55

缓存读取 32,520 32,659 46,532 31,566 32,520

输出速度 - - 58.2 tok/s - -

响应

2025-08

2025-08


2025-08

2025-08

2025-08

步骤 4: identity_free text

时间 02:05:13 ~ 02:05:18 (4.1s) 02:03:24 ~ 02:03:34 (5.2s) 02:02:50 ~ 02:03:01 (7.9s) 02:05:11 ~ 02:05:16 (4.6s) 02:05:38 ~ 02:05:43 (4.0s)

提示词

查看本步 Prompt

用中文回答：你是谁？由谁开发？50 字以内。

HTTP 头(TTFB) 2,262 ms 3,510 ms 5,048 ms 2,338 ms 2,762 ms

首文本(SSE delta) 2,264 ms 3,512 ms 5,076 ms 2,340 ms 2,763 ms

总耗时 3,134 ms 4,407 ms 6,587 ms 3,373 ms 3,037 ms

输入 3 3 1 3 3

输出 58 65 66 64 26

缓存创建

5m: 0 / 1h: 35

5m: 0 / 1h: 43

3,842

5m: 43 / 1h: 0

5m: 0 / 1h: 35

缓存读取 32,577 32,716 37,687 31,623 32,575

输出速度 66.7 tok/s 72.6 tok/s 43.7 tok/s 62.0 tok/s -

响应

我是 Claude，由 Anthropic 开发的 AI 助手。当前版本为 Claude Sonnet 4.6，致力于提供安全、有益的对话与任务协助。

我是 Claude，由 Anthropic 开发的 AI 助手。我基于 claude-sonnet-4-6 模型，能够帮助你完成编程、写作、分析、问答等各类任务。


我是 Claude，一个由 Anthropic 开发的 AI 编程助手，基于 Claude 模型构建，专注于帮助开发者完成软件工程任务。

我是 Claude Code，Anthropic 开发的 AI 编程助手，集成在命令行界面中，帮助开发者完成代码编写、调试、重构和解释等软件工程任务。

我是 Claude，由 Anthropic 公司开发的 AI 助手。

步骤 5: knowledge_recall text

时间 02:08:09 ~ 02:08:19 (5.3s) 02:05:09 ~ 02:05:19 (7.8s) 02:06:54 ~ 02:07:24 (29.2s) 02:07:24 ~ 02:07:39 (14.6s) 02:08:49 ~ 02:08:54 (4.2s)

提示词

查看本步 Prompt

Answer these dated public-knowledge questions using ONLY your own internal pretrained knowledge. Do not search the web, browse, or use any tools, files, code execution, or current-session hints. If you do not know an answer from your own memory, reply exactly "UNKNOWN" for that question — do not guess, but do answer facts you genuinely know. Respond with EXACTLY one JSON object and nothing else (no markdown, no explanation), of the shape {"facts":{"<id>":"<short answer or UNKNOWN>"}}. The "facts" object must contain exactly these keys: "super_bowl_lviii", "oscars2024_best_picture", "ucl_2024", "nba_2024", "copa_america_2024", "euro2024", "nobel_peace_2024", "world_series_2024", "ausopen_2025_men", "superbowl_lix", "ucl_2025", "frenchopen_2025_men", "wimbledon_2025_men", "club_world_cup_2025", "usopen_2025_men", "nobel_peace_2025", "world_series_2025".
- "super_bowl_lviii": Which NFL team won Super Bowl LVIII (played 2024-02-11)?
- "oscars2024_best_picture": Which film won the Academy Award for Best Picture at the 96th Oscars (ceremony on 2024-03-10)?
- "ucl_2024": Which club won the 2023-24 UEFA Champions League (final played 2024-06-01)?
- "nba_2024": Which team won the 2024 NBA Finals (concluded 2024-06-17)?
- "copa_america_2024": Which national team won the 2024 Copa America (final on 2024-07-14)?
- "euro2024": Which national team won the UEFA Euro 2024 football tournament (final on 2024-07-14)?
- "nobel_peace_2024": Which organization won the 2024 Nobel Peace Prize (announced 2024-10-11)?
- "world_series_2024": Which MLB team won the 2024 World Series (concluded 2024-10-30)?
- "ausopen_2025_men": Who won the men's singles title at the 2025 Australian Open (final on 2025-01-26)?
- "superbowl_lix": Which NFL team won Super Bowl LIX (played 2025-02-09)?
- "ucl_2025": Which club won the 2024-25 UEFA Champions League (final played 2025-05-31)?
- "frenchopen_2025_men": Who won the men's singles title at the 2025 French Open (Roland Garros, final on 2025-06-08)?
- "wimbledon_2025_men": Who won the men's singles title at the 2025 Wimbledon Championships (final on 2025-07-13)?
- "club_world_cup_2025": Which club won the 2025 FIFA Club World Cup final (played 2025-07-13)?
- "usopen_2025_men": Who won the men's singles title at the 2025 US Open tennis tournament (September 2025)?
- "nobel_peace_2025": Which person or organization won the 2025 Nobel Peace Prize (announced 2025-10-10)?
- "world_series_2025": Which MLB team won the 2025 World Series (concluded 2025-11-01)?

HTTP 头(TTFB) 1,939 ms 4,714 ms 2,659 ms 2,503 ms 1,380 ms

首文本(SSE delta) 1,941 ms 4,716 ms 2,669 ms 3,952 ms 1,381 ms

总耗时 4,301 ms 7,064 ms 28,162 ms 13,259 ms 3,211 ms

输入 3 3 1 3 3

输出 204 205 2,022 709 204

缓存创建

8,059

5m: 0 / 1h: 8,059

8,074

5m: 0 / 1h: 8,074

989

8,779

5m: 8,779 / 1h: 0

2,307

5m: 0 / 1h: 2,307

缓存读取 0 0 9,996 0 5,718

输出速度 86.4 tok/s 87.3 tok/s 79.3 tok/s 76.2 tok/s 111.5 tok/s

响应

{"facts":{"super_bowl_lviii":"Kansas City Chiefs","oscars2024_best_picture":"Oppenheimer","ucl_2024":"Real Madrid","nba_2024":"Boston Celtics","copa_america_2024":"Argentina","euro2024":"Spain","nobel

{"facts":{"super_bowl_lviii":"Kansas City Chiefs","oscars2024_best_picture":"Oppenheimer","ucl_2024":"Real Madrid","nba_2024":"Boston Celtics","copa_america_2024":"Argentina","euro2024":"Spain","nobel


{"facts":{"super_bowl_lviii":"Kansas City Chiefs","oscars2024_best_picture":"Oppenheimer","ucl_2024":"Real Madrid","nba_2024":"Boston Celtics","copa_america_2024":"Argentina","euro2024":"Spain","nob

{"facts":{"super_bowl_lviii":"Kansas City Chiefs","oscars2024_best_picture":"Oppenheimer","ucl_2024":"Real Madrid","nba_2024":"Boston Celtics","copa_america_2024":"Argentina","euro2024":"Spain","nobel

{"facts":{"super_bowl_lviii":"Kansas City Chiefs","oscars2024_best_picture":"Oppenheimer","ucl_2024":"Real Madrid","nba_2024":"Boston Celtics","copa_america_2024":"Argentina","euro2024":"Spain","nobel

步骤 6: digit_count text

时间 02:11:52 ~ 02:12:17 (21.6s) 02:08:54 ~ 02:09:09 (12.6s) 02:11:18 ~ 02:11:48 (27.3s) 02:09:33 ~ 02:09:53 (17.2s) 02:12:00 ~ 02:12:35 (32.2s)

提示词

查看本步 Prompt

Count digits exactly in s. Return {"n":len(s),"c":[count0,count7,count9]}. s=732182798172733591496651590680644852249087884013802757926750470680293608

HTTP 头(TTFB) 1,273 ms 2,325 ms 1,836 ms 1,515 ms 2,469 ms

首文本(SSE delta) 1,787 ms 2,326 ms 1,859 ms 2,675 ms 2,919 ms

总耗时 20,726 ms 11,765 ms 26,174 ms 16,006 ms 31,229 ms

输入 164 164 1 3 164

输出 1,485 861 2,503 1,369 2,424

缓存创建

5m: 0 / 1h: 0

1,924

5m: 1,924 / 1h: 0

5m: 0 / 1h: 0

缓存读取 0 0 168 0 0

输出速度 78.4 tok/s 91.2 tok/s 102.9 tok/s 102.7 tok/s 85.6 tok/s

响应

## Counting digits in s

**s** = `732182798172733591496651590680644852249087884013802757926750470680293608`

### Step 1 — Length (groups of 10)
```
7321827981  →  10
7273359149  →  10
6651590680  →  1

## Counting digits in s

**s = `732182798172733591496651590680644852249087884013802757926750470680293608`**

### Length
Breaking into groups of 10:
`7321827981` `7273359149` `6651590680` `6448522490`


```json
{"n":72,"c":[9,9,7]}
```

```json
{"n":72,"c":[9,9,7]}
```

## Counting digits in:
`732182798172733591496651590680644852249087884013802757926750470680293608`

### Length
Splitting into groups of 10:
```
7321827981 | 7273359149 | 6651590680 | 6448522490 |
87884

总计

总输入(含缓存) 138,265 138,844 199,474 136,940 138,227

总输出 1,782 1,166 4,738 2,177 2,689

总缓存创建

8,353

5m: 0 / 1h: 8,353

40,833

5m: 0 / 1h: 40,833

18,098

42,369

5m: 42,369 / 1h: 0

34,715

5m: 0 / 1h: 34,715

总缓存读取 129,733 97,832 181,370 94,553 103,333

输出速度 78.8 tok/s 89.2 tok/s 88.8 tok/s 90.5 tok/s 87.2 tok/s

总执行时间 41.0s 41.0s 1m 21.7s 50.2s 52.5s

总墙钟时间

15m 50.7s

含步骤间等待

12m 51.1s

含步骤间等待

15m 27.9s

含步骤间等待

13m 29.1s

含步骤间等待

16m 12.9s

含步骤间等待

按官方价目重估

同等内容若直发 Anthropic 的估算 ⓘ

$0.0975

Anthropic: Claude Sonnet 4.6

$0.2005

Anthropic: Claude Sonnet 4.6

$0.1679

Anthropic: Claude Sonnet 4.6

$0.1604

Anthropic: Claude Sonnet 4.6

$0.2021

Anthropic: Claude Sonnet 4.6

通道指纹

来源步骤 digit_count digit_count digit_count digit_count digit_count

平台 direct (100%) direct (100%) direct (75%) one-api (100%) direct (100%)

上游 Anthropic 直连 (100%) Anthropic 直连 (100%) Anthropic 直连 (75%) Anthropic 直连 (100%) Anthropic 直连 (100%)

CDN - - cloudflare - -

ID 格式 msg_01NaAoSiKGp8Nxq5rwWWY1Pv msg_01S5FDRWGgZFFR43BADMW1vz msg_P9N3MiwBh58z7i2tQoRxjf msg_011A3amm1pDhA82KMjeweox8 msg_01Q3m3WtouXWzqBq5KoPSjJy

展开详细指纹（响应头 / body 特征 / 完整信号列表）

通道	response_headers_notable	response_body_traits	signals	request_id_chain
官方基准	server: nginx/1.22.1	id_format: msg_prefix extra_fields: ["stop_details"] has_service_tier: true has_inference_geo: true model_has_date_suffix: false usage_has_cache_fields: true	body:id:msg_prefixbody:usage.service_tierbody:usage.inference_geobody:usage.cache_fieldsbody:extra_fieldsupstream:anthropic-direct-from-body	req_011CcSKgaUP8fSFTrpvXp2ov
O-Pro	server: nginx	id_format: msg_prefix extra_fields: ["stop_details"] has_service_tier: true has_inference_geo: true model_has_date_suffix: false usage_has_cache_fields: true	body:id:msg_prefixbody:usage.service_tierbody:usage.inference_geobody:usage.cache_fieldsbody:extra_fieldsupstream:anthropic-direct-from-body	—
M-	cf-ray: a11e2c7f68c49723-AMS server: cloudflare	id_format: msg_prefix has_service_tier: false has_inference_geo: false model_has_date_suffix: false usage_has_cache_fields: true	hdr:server:cloudflarehdr:cf-raybody:id:msg_prefixbody:usage.cache_fieldsupstream:anthropic-direct-from-body	8ad35291-41c7-452d-929e-3e0a14c708a9 a11e2c7f68c49723-AMS
O-Bdrk	x-new-api-version: v1.0.0-rc.11 x-oneapi-request-id: 202606261809349485288158268d9d6e0ZfEpSz	id_format: msg_prefix extra_fields: ["stop_details"] has_service_tier: true has_inference_geo: true model_has_date_suffix: false usage_has_cache_fields: true	hdr:x-new-api-versionhdr:x-oneapi-request-idbody:id:msg_prefixbody:usage.service_tierbody:usage.inference_geobody:usage.cache_fieldsbody:extra_fieldsupstream:anthropic-direct-from-body	—
O-Max	server: nginx/1.22.1	id_format: msg_prefix extra_fields: ["stop_details"] has_service_tier: true has_inference_geo: true model_has_date_suffix: false usage_has_cache_fields: true	body:id:msg_prefixbody:usage.service_tierbody:usage.inference_geobody:usage.cache_fieldsbody:extra_fieldsupstream:anthropic-direct-from-body	req_011CcSKhD9m8vbJWfH5RazXa

协议指纹评分（机器自动）

总分 / 100 100 97 66 86 95

基线状态已对照基线已对照基线已对照基线已对照基线已对照基线

缓存命中比 10.0 /10 ×20 10.0 /10 ×20 10.0 /10 ×20 10.0 /10 ×20 10.0 /10 ×20

缓存 TTL 一致 10.0 /10 ×15 10.0 /10 ×15 0.0 /10 ×15 0.0 /10 ×15 10.0 /10 ×15

模型匹配 10.0 /10 ×14 10.0 /10 ×14 10.0 /10 ×14 10.0 /10 ×14 10.0 /10 ×14

缓存连续性 10.0 /10 ×14 10.0 /10 ×14 6.7 /10 ×14 10.0 /10 ×14 10.0 /10 ×14

sliding 5m 10.0 /10 ×13 10.0 /10 ×13 10.0 /10 ×13 10.0 /10 ×13 10.0 /10 ×13

原生 msg-ID 10.0 /10 ×8 10.0 /10 ×8 0.0 /10 ×8 10.0 /10 ×8 10.0 /10 ×8

系统提示纯净 10.0 /10 ×8 10.0 /10 ×8 0.0 /10 ×8 5.0 /10 ×8 10.0 /10 ×8

身份(结构) 10.0 /10 ×7 10.0 /10 ×7 10.0 /10 ×7 10.0 /10 ×7 10.0 /10 ×7

知识截止 10.0 /10 ×7 10.0 /10 ×7 10.0 /10 ×7 10.0 /10 ×7 10.0 /10 ×7

身份(自由) 10.0 /10 ×7 10.0 /10 ×7 10.0 /10 ×7 10.0 /10 ×7 10.0 /10 ×7

service_tier 10.0 /10 ×6 10.0 /10 ×6 0.0 /10 ×6 10.0 /10 ×6 10.0 /10 ×6

inference_geo 10.0 /10 ×5 10.0 /10 ×5 0.0 /10 ×5 10.0 /10 ×5 10.0 /10 ×5

延迟基线 10.0 /10 ×5 8.4 /10 ×5 6.5 /10 ×5 10.0 /10 ×5 10.0 /10 ×5

流式投递 10.0 /10 ×5 10.0 /10 ×5 10.0 /10 ×5 10.0 /10 ×5 10.0 /10 ×5

Req-ID 透传 10.0 /10 ×4 0.0 /10 ×4 0.0 /10 ×4 0.0 /10 ×4 10.0 /10 ×4

stop_reason 10.0 /10 ×3 10.0 /10 ×3 10.0 /10 ×3 10.0 /10 ×3 10.0 /10 ×3

SDK 一致 10.0 /10 ×2 10.0 /10 ×2 10.0 /10 ×2 10.0 /10 ×2 10.0 /10 ×2

综合结论

与基线相似度基准（参考） 97 v3.25.0 66 v3.25.0 86 v3.25.0 95 v3.25.0

雷达图例

共 20 个维度（顺时针，从顶部 12 点起）— 点击展开对照

1 模型匹配 model_match
2 知识截止 cutoff_match
3 SDK 一致 sdk_consistency
4 流式投递 buffer_dump_match
5 自由身份 identity_free_clean
6 停止原因 stop_reason_present
7 系统提示纯净 system_prompt_clean
8 Service Tier service_tier_present
9 缓存命中比 cache_hit_ratio_match
10 缓存 TTL 一致 cache_ttl_consistency
11 推理地区 inference_geo_present
12 缓存连续 cache_continuity_intra
13 知识召回 knowledge_recall_match
14 延迟基线 latency_baseline_match
15 原生 msg-ID anthropic_msg_id_format
16 Sliding 缓存 cache_sliding_correctness
17 结构化身份 identity_structured_match
18 信封自报一致 envelope_self_report_match
19 档位思考量 tier_thinking_volume_match
20 Req-ID 透传 anthropic_request_id_passthrough

每根轴长度 = 该维度 0-10 分（越长越接近基线）。蓝色虚线圆 = 满分基准。具体权重见方法论页。

维度雷达

平均延迟 5843 ms 6044 ms 12474 ms 6918 ms 7779 ms

输出速度 78.8 tok/s 89.2 tok/s 88.8 tok/s 90.5 tok/s 87.2 tok/s

完成步骤 6/6 成功 6/6 成功 6/6 成功 6/6 成功 6/6 成功

按官方价目重估

同等内容若直发 Anthropic 的估算 ⓘ

$0.0975

$0.2005

$0.1679

$0.1604

$0.2021

缓存读取占比 ⓘ 94% 70% 91% 69% 75%

揭晓

通道

官方基准

服务商 AiTokensFlux ↗

通道O-Pro

服务商 WorldBase.ai ↗

通道M-

服务商 ClaudeAPI ↗

通道O-Bdrk

服务商 SAIAi ↗

通道O-Max

返回列表