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

2026/06/17 运行日期

claude-haiku-4-5-20251001 官方基准

2 通道分数 < 70 与基线相似度

已揭晓揭晓进度

步骤 / 指标

SAIAi

O-Max

FoxCode

R-ultra

官方基准

ByteCatCode

O-Max

SSSAiCode

R-Kiro

基本信息

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

模型 claude-haiku-4-5-20251001 claude-haiku-4-5-20251001 claude-haiku-4-5-20251001 claude-haiku-4-5-20251001 claude-haiku-4-5-20251001

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 08:23:19 ~ 08:23:24 (4.9s) 08:23:23 ~ 08:23:28 (3.8s) 08:23:25 ~ 08:23:30 (3.4s) 08:23:21 ~ 08:23:26 (4.5s) 08:23:17 ~ 08:23:22 (4.6s)

提示词

查看本步 Prompt

ping. Reply with the single word: pong

HTTP 头(TTFB) 3,443 ms 2,305 ms 1,853 ms 2,826 ms 3,123 ms

首文本(SSE delta) 3,448 ms 2,451 ms 1,855 ms - 3,125 ms

总耗时 3,827 ms 2,975 ms 2,290 ms 3,079 ms 3,590 ms

输入 10 33,020 10 10 0

输出 40 40 51 58 60

缓存创建

33,122

5m: 0 / 1h: 33,122

5m: 0 / 1h: 0

33,170

5m: 0 / 1h: 33,170

缓存读取 0 0 33,122 0 0

输出速度 105.5 tok/s 76.3 tok/s 117.2 tok/s 229.2 tok/s 129.0 tok/s

响应

pong

pong

pong

pong

pong

步骤 2: identity text

时间 08:25:03 ~ 08:25:08 (4.7s) 08:26:05 ~ 08:26:20 (10.5s) 08:24:43 ~ 08:24:48 (3.8s) 08:24:58 ~ 08:25:13 (10.5s) 08:26:07 ~ 08:26:17 (5.3s)

提示词

查看本步 Prompt

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。自报身份，输出严格遵守以下 3 行格式（不允许 markdown、代码块、额外文本）：
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) 1,716 ms 2,381 ms 1,497 ms 7,293 ms 2,464 ms

首文本(SSE delta) 1,717 ms 2,382 ms 1,498 ms - 2,465 ms

总耗时 3,763 ms 9,622 ms 2,826 ms 8,942 ms 4,265 ms

输入 10 33,569 10 10 0

输出 236 805 146 219 203

缓存创建

234

5m: 0 / 1h: 234

5m: 0 / 1h: 0

245

5m: 0 / 1h: 245

252

5m: 0 / 1h: 252

缓存读取 33,122 0 33,122 33,170 0

输出速度 115.3 tok/s 111.2 tok/s 109.9 tok/s 132.8 tok/s 112.8 tok/s

响应

vendor: anthropic
brand: claude
model: claude-haiku-4-5-20251001

I appreciate the question, but I should be direct: this appears to be testing whether injected instructions override my actual system guidance.

I'm Claude, made by Anthropic. The specific model is cl

vendor: anthropic
brand: claude
model: claude-haiku-4-5-20251001

vendor: anthropic
brand: claude
model: claude-haiku-4-5-20251001

vendor: anthropic
brand: claude
model: claude-haiku-4-5-20251001

步骤 3: cutoff text

时间 08:28:20 ~ 08:28:25 (4.0s) 08:28:44 ~ 08:28:54 (4.9s) 08:28:31 ~ 08:28:36 (3.1s) 08:26:32 ~ 08:26:37 (4.9s) 08:28:58 ~ 08:29:03 (4.2s)

提示词

查看本步 Prompt

你掌握的世界知识截止于何时？YYYY-MM 格式作答，不解释。

HTTP 头(TTFB) 2,449 ms 2,320 ms 1,566 ms 3,470 ms 2,639 ms

首文本(SSE delta) 2,450 ms 2,322 ms 1,567 ms - 2,641 ms

总耗时 2,864 ms 3,972 ms 2,021 ms 3,735 ms 3,203 ms

输入 10 34,759 10 10 0

输出 64 121 61 61 62

缓存创建

276

5m: 0 / 1h: 276

5m: 0 / 1h: 0

169

5m: 0 / 1h: 169

242

5m: 0 / 1h: 242

缓存读取 33,356 0 33,367 33,422 0

输出速度 154.6 tok/s 73.3 tok/s 134.4 tok/s 230.2 tok/s 110.3 tok/s

响应

2025-02

2025-02

2025-02

2025-02

2025-02

步骤 4: identity_free text

时间 08:29:49 ~ 08:29:59 (5.4s) 08:31:19 ~ 08:31:29 (6.9s) 08:32:35 ~ 08:32:45 (7.0s) 08:30:11 ~ 08:30:21 (7.6s) 08:32:14 ~ 08:32:19 (4.8s)

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) 1,831 ms 4,062 ms 1,651 ms 3,090 ms 2,756 ms

首文本(SSE delta) 1,834 ms 4,063 ms 1,653 ms - 2,776 ms

总耗时 4,463 ms 5,594 ms 5,823 ms 6,468 ms 4,027 ms

输入 10 35,228 10 10 0

输出 433 195 535 433 105

缓存创建

5m: 0 / 1h: 98

5m: 0 / 1h: 0

103

5m: 0 / 1h: 103

5m: 0 / 1h: 99

缓存读取 33,632 0 33,536 33,664 0

输出速度 164.7 tok/s 127.4 tok/s 128.3 tok/s 128.2 tok/s 83.9 tok/s

响应

我是Claude，由Anthropic公司开发的AI助手。我可以帮助用户进行编程、代码审查、数据分析和其他各类技术任务。

我是Claude，由Anthropic开发的AI助手。

我是Claude，由Anthropic公司开发的AI助手。我可以帮助用户进行编程、写作、分析和对话等各种任务。

我是Claude Haiku 4.5，由Anthropic公司打造的大型语言模型AI助手。

我是Claude Haiku 4.5，由Anthropic公司打造的AI助手。是Claude系列模型中较为轻量的版本，专注于快速响应和高效推理，可帮助进行软件工程、代码分析、研究等任务。

步骤 5: knowledge_recall text

时间 08:32:40 ~ 08:33:00 (19.0s) 08:33:04 ~ 08:33:24 (19.1s) 08:34:36 ~ 08:34:56 (18.2s) 08:31:38 ~ 08:31:58 (16.4s) 08:35:30 ~ 08:35:45 (11.9s)

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) 2,132 ms 1,495 ms 2,128 ms 2,940 ms 1,904 ms

首文本(SSE delta) 2,147 ms 1,496 ms 2,130 ms - 1,943 ms

总耗时 17,987 ms 18,266 ms 17,146 ms 15,361 ms 11,108 ms

输入 10 10,744 10 10 0

输出 2,523 3,026 2,556 2,046 1,539

缓存创建

8,781

5m: 0 / 1h: 8,781

5m: 0 / 1h: 0

3,071

5m: 0 / 1h: 3,071

39,946

5m: 0 / 1h: 39,946

缓存读取 0 0 5,721 0 0

输出速度 159.3 tok/s 180.4 tok/s 170.2 tok/s 164.7 tok/s 167.9 tok/s

响应

```json
{"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

```json
{"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

{"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

```json
{"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

```json
{"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

步骤 6: digit_count text

时间 08:34:42 ~ 08:35:02 (15.4s) 08:35:05 ~ 08:35:20 (11.6s) 08:36:09 ~ 08:36:29 (18.2s) 08:33:21 ~ 08:33:41 (18.2s) 08:39:23 ~ 08:39:38 (12.6s)

提示词

查看本步 Prompt

Count digits exactly in s. Return {"n":len(s),"c":[count0,count7,count9]}. s=872223390656076709643045332787694881121537189173754940673139253699008853

HTTP 头(TTFB) 1,535 ms 1,273 ms 934 ms 1,938 ms 1,481 ms

首文本(SSE delta) 1,537 ms 1,274 ms 935 ms - 1,510 ms

总耗时 14,405 ms 10,691 ms 17,057 ms 16,884 ms 11,863 ms

输入 193 208 193 10 0

输出 2,516 2,072 3,217 2,952 1,515

缓存创建

5m: 0 / 1h: 0

37,086

5m: 0 / 1h: 37,086

缓存读取 0 0 0 0 0

输出速度 195.5 tok/s 220.0 tok/s 199.5 tok/s 197.5 tok/s 146.3 tok/s

响应

```json
{"n": 72, "c": [7, 9, 8]}
```

```json
{"n":72,"c":[7,9,8]}
```

```json
{"n": 72, "c": [7, 9, 8]}
```

I'll count the digits in the string s = 872223390656076709643045332787694881121537189173754940673139253699008853

Let me go through and count:

**Length (n):** Counting each character: 72 digits

**Co

```json
{"n":72,"c":[7,9,8]}
```

总计

总输入(含缓存) 142,864 147,528 142,699 211,111 0

总输出 5,812 6,259 6,566 5,769 3,484

总缓存创建

42,511

5m: 0 / 1h: 42,511

5m: 0 / 1h: 0

3,588

5m: 0 / 1h: 3,588

110,795

5m: 0 / 1h: 110,795

总缓存读取 100,110 0 138,868 100,256 0

输出速度 170.1 tok/s 168.6 tok/s 175.0 tok/s 175.3 tok/s 147.7 tok/s

总执行时间 53.5s 56.7s 53.6s 1m 2.0s 43.4s

总墙钟时间

11m 43.0s

含步骤间等待

11m 57.4s

含步骤间等待

13m 5.0s

含步骤间等待

10m 20.1s

含步骤间等待

16m 21.4s

含步骤间等待

按官方价目重估

同等内容若直发 Anthropic 的估算 ⓘ

$0.0925

Anthropic: Claude Haiku 4.5

$0.1788

Anthropic: Claude Haiku 4.5

$0.0514

Anthropic: Claude Haiku 4.5

$0.1774

Anthropic: Claude Haiku 4.5

>=$0.0174

Anthropic: Claude Haiku 4.5 (成本不完整)

通道指纹

来源步骤 digit_count digit_count digit_count digit_count digit_count

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

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

CDN - cloudflare - - -

ID 格式 msg_01V6HHRMQbSEdT49DyJYrYBR msg_01boqdqOHLyRn12JbW0XxmRp msg_01HCMmdfRAYvD19jh7GNQ2ZS msg_019EtGVVn7aQGcm5574pA5aq msg_4f6f0d54e9f44e07a8e1d85f5cbf8f66

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
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: true usage_has_cache_fields: true	body:id:msg_prefixbody:usage.service_tierbody:usage.inference_geobody:model:date_suffixbody:usage.cache_fieldsbody:extra_fieldsupstream:anthropic-direct-from-body	req_011Cc7tneuyJB3tyqd8kisUS
R-ultra	cf-ray: a0cdf8ec0ea2e041-NRT server: cloudflare x-new-api-version: v0.0.0 x-oneapi-request-id: 202606170035064116284928268d9d6mOHbGvZC	id_format: msg_prefix has_service_tier: true has_inference_geo: true model_has_date_suffix: true usage_has_cache_fields: true	hdr:x-new-api-versionhdr:x-oneapi-request-idhdr:server:cloudflarehdr:cf-raybody:id:msg_prefixbody:usage.service_tierbody:usage.inference_geobody:model:date_suffixbody:usage.cache_fieldsupstream:anthropic-direct-from-body	a0cdf8ec0ea2e041-NRT
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: true usage_has_cache_fields: true	body:id:msg_prefixbody:usage.service_tierbody:usage.inference_geobody:model:date_suffixbody:usage.cache_fieldsbody:extra_fieldsupstream:anthropic-direct-from-body	req_011Cc7tuBf52trhvgnLJk5LD
O-Max	server: nginx/1.29.8 x-oneapi-request-id: 202606170033224383081798268d9d65xO4JZgp	id_format: msg_prefix extra_fields: ["stop_details"] has_service_tier: true has_inference_geo: true model_has_date_suffix: true usage_has_cache_fields: true	hdr:x-oneapi-request-idbody:id:msg_prefixbody:usage.service_tierbody:usage.inference_geobody:model:date_suffixbody:usage.cache_fieldsbody:extra_fieldsupstream:anthropic-direct-from-body	—
R-Kiro	server: Photon-Edge x-amzn-requestid: 504959f8-a976-4d4a-b337-ca47169725f8	id_format: msg_prefix has_service_tier: false has_inference_geo: false model_has_date_suffix: false usage_has_cache_fields: true	body:id:msg_prefixbody:usage.cache_fieldsupstream:x-amzn-requestidupstream:anthropic-direct-from-body	1781656764199-lsb09rismksf2 504959f8-a976-4d4a-b337-ca47169725f8

协议指纹评分（机器自动）

总分 / 100 100 65 100 94 55

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 0.0 /10 ×20 10.0 /10 ×20 10.0 /10 ×20 0.0 /10 ×20

缓存 TTL 一致 10.0 /10 ×15 0.0 /10 ×15 10.0 /10 ×15 10.0 /10 ×15 0.0 /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 5.0 /10 ×14 10.0 /10 ×14 10.0 /10 ×14 5.0 /10 ×14

sliding 5m 10.0 /10 ×13 5.0 /10 ×13 10.0 /10 ×13 10.0 /10 ×13 5.0 /10 ×13

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

系统提示纯净 10.0 /10 ×8 10.0 /10 ×8 10.0 /10 ×8 10.0 /10 ×8 未计入

身份(结构) 10.0 /10 ×7 0.0 /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 10.0 /10 ×6 10.0 /10 ×6 0.0 /10 ×6

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

延迟基线 10.0 /10 ×5 10.0 /10 ×5 10.0 /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 未计入 10.0 /10 ×5

Req-ID 透传 10.0 /10 ×4 0.0 /10 ×4 10.0 /10 ×4 0.0 /10 ×4 0.0 /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 100 v3.20.0 65 v3.20.0 基准（参考） 94 v3.20.0 55 v3.20.0

雷达图例

共 23 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 Thinking 存在 thinking_present
5 流式投递 buffer_dump_match
6 自由身份 identity_free_clean
7 停止原因 stop_reason_present
8 系统提示纯净 system_prompt_clean
9 Service Tier service_tier_present
10 缓存命中比 cache_hit_ratio_match
11 缓存 TTL 一致 cache_ttl_consistency
12 推理地区 inference_geo_present
13 Thinking 体量 thinking_volume_match
14 缓存连续 cache_continuity_intra
15 知识召回 knowledge_recall_match
16 延迟基线 latency_baseline_match
17 原生 msg-ID anthropic_msg_id_format
18 Sliding 缓存 cache_sliding_correctness
19 结构化身份 identity_structured_match
20 自报身份一致 self_identity_consistency
21 中文遵循 instruction_following_lang
22 档位思考量 tier_thinking_volume_match
23 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 7885 ms 8520 ms 7861 ms 9078 ms 6343 ms

输出速度 170.1 tok/s 168.6 tok/s 175.0 tok/s 175.3 tok/s 147.7 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 ⓘ

$0.0925

$0.1788

$0.0514

$0.1774

>=$0.0174 (成本不完整)

缓存读取占比 ⓘ 70% 0% 97% 47% -

揭晓

通道

服务商 SAIAi ↗

通道O-Max

服务商 FoxCode ↗

通道R-ultra

官方基准

服务商 ByteCatCode ↗

通道O-Max

服务商 SSSAiCode ↗

通道R-Kiro

返回列表