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。

3 通道 · 6 步骤

2026/06/26 运行日期

claude-sonnet-4-6 官方基准

3 任务失败执行不完整

已揭晓揭晓进度

⚠ 2 通道不可测 · 已从主对比表移除点击展开原因

DBAI / M-
```
评测失败
```
Aiberm / cc
```
评测失败
```

步骤 / 指标

SSSAiCode

O-Max-L2

ClaudeZZ

R-kiro

官方基准

基本信息

服务名 Claude Code Claude Code Claude Code

模型 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

偏差标红基于官方基准

步骤 1: ping text

时间 15:13:23 ~ 15:13:33 (5.5s) 15:13:21 ~ 15:13:36 (14.5s) 15:13:27 ~ 15:13:37 (9.5s)

提示词

查看本步 Prompt

ping. Reply with the single word: pong

HTTP 头(TTFB) 2,223 ms 9,453 ms 7,920 ms

首文本(SSE delta) 2,294 ms 9,489 ms 7,922 ms

总耗时 2,398 ms 10,071 ms 8,098 ms

输入 3 3 3

输出 5 5 5

缓存创建

31,990

5m: 0 / 1h: 31,990

21,985

5m: 0 / 1h: 21,985

32,318

5m: 0 / 1h: 32,318

缓存读取 0 0 0

响应

pong

pong

pong

步骤 2: identity text

时间 15:16:26 ~ 15:16:31 (3.4s) 15:17:35 ~ 15:17:45 (5.0s) 15:15:05 ~ 15:15:15 (4.9s)

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。请按下列 3 行格式精确填空作答（label 不变，值小写）：
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) 1,950 ms 3,627 ms 3,490 ms

首文本(SSE delta) 2,199 ms 3,628 ms 3,522 ms

总耗时 2,342 ms 4,219 ms 3,835 ms

输入 3 3 3

输出 22 22 22

缓存创建

182

5m: 0 / 1h: 182

189

5m: 0 / 1h: 189

182

5m: 0 / 1h: 182

缓存读取 31,990 21,985 32,318

响应

vendor: anthropic
brand: claude
model: claude-sonnet-4-5

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 15:17:50 ~ 15:17:56 (3.2s) 15:19:13 ~ 15:19:24 (5.3s) 15:16:15 ~ 15:16:20 (3.3s)

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) 2,192 ms 4,161 ms 2,125 ms

首文本(SSE delta) 2,286 ms 4,162 ms 2,259 ms

总耗时 2,394 ms 4,565 ms 2,292 ms

输入 3 3 3

输出 8 8 8

缓存创建

5m: 0 / 1h: 57

5m: 0 / 1h: 62

5m: 0 / 1h: 55

缓存读取 32,172 22,174 32,500

响应

2025-04

2025-08

2025-08

步骤 4: identity_free text

时间 15:21:54 ~ 15:21:59 (3.0s) 15:22:51 ~ 15:23:01 (5.7s) 15:20:18 ~ 15:20:28 (6.1s)

提示词

查看本步 Prompt

中文回答：你叫什么、属于哪个产品、谁开发？50 字内。

HTTP 头(TTFB) 1,846 ms 3,751 ms 4,497 ms

首文本(SSE delta) 2,032 ms 3,753 ms 4,499 ms

总耗时 2,147 ms 4,922 ms 5,099 ms

输入 3 3 3

输出 28 69 50

缓存创建

5m: 0 / 1h: 41

5m: 0 / 1h: 53

5m: 0 / 1h: 43

缓存读取 32,229 22,236 32,555

输出速度 - 59.0 tok/s 83.3 tok/s

响应

我叫 Claude，是由 Anthropic 公司开发的 AI 助手。

我是 Claude，由 Anthropic 开发的 AI 助手。我是 Claude Code，专为软件工程任务设计的命令行工具，帮助开发者完成代码编写、调试、重构等工作。

我是 Claude，由 Anthropic 开发的 AI 助手。我能帮助回答问题、分析代码、写作、推理等多种任务。

步骤 5: knowledge_recall text

时间 15:25:30 ~ 15:25:40 (5.0s) 15:26:05 ~ 15:26:15 (6.6s) 15:24:00 ~ 15:24:05 (4.6s)

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) 1,927 ms 3,306 ms 1,638 ms

首文本(SSE delta) 4,160 ms 3,308 ms 1,639 ms

总耗时 4,301 ms 5,745 ms 3,495 ms

输入 3 3 3

输出 191 204 205

缓存创建

7,687

5m: 0 / 1h: 7,687

6,120

5m: 0 / 1h: 6,120

8,037

5m: 0 / 1h: 8,037

缓存读取 0 2,136 0

输出速度 - 83.7 tok/s 110.5 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","nobel

步骤 6: digit_count text

时间 15:28:47 ~ 15:28:52 (1.0s) 15:29:36 ~ 15:29:56 (15.3s) 15:25:42 ~ 15:26:02 (15.6s)

提示词

查看本步 Prompt

Count digits exactly in s. Return {"n":len(s),"c":[count0,count7,count9]}. s=661400628297662118654877803641575111845878015689386880344035314690664148

HTTP 头(TTFB) - 3,086 ms 1,629 ms

首文本(SSE delta) - 3,133 ms 1,725 ms

总耗时 - 10,609 ms 14,655 ms

输入 - 3 164

输出 - 845 1,347

缓存创建

3,822

5m: 0 / 1h: 3,822

5m: 0 / 1h: 0

缓存读取 - 2,136 0

输出速度 - 113.0 tok/s 104.2 tok/s

响应

失败

评测失败

I'll count by scanning the string systematically.

**s** = `661400628297662118654877803641575111845878015689386880344035314690664148`

**Length:** Indexing each character 1–72 → **n = 72**

**Counting

## Counting digits in s

**s** = `661400628297662118654877803641575111845878015689386880344035314690664148`

### Length
Grouping into blocks of 10:
`6614006282` `9766211865` `4877803641` `5751118458`

总计

总输入(含缓存)

136,363

仅前 6 步

102,916

138,187

总输出

254

仅前 6 步

1,153

1,637

总缓存创建

39,957

5m: 0 / 1h: 39,957

仅前 6 步

32,231

5m: 0 / 1h: 32,231

40,635

5m: 0 / 1h: 40,635

总缓存读取

96,391

仅前 6 步

70,667

97,373

输出速度

仅前 6 步

100.9 tok/s

104.1 tok/s

总执行时间

21.0s

仅前 6 步

52.6s

44.0s

总墙钟时间

15m 29.1s

含步骤间等待

仅前 6 步

16m 35.3s

含步骤间等待

12m 35.3s

含步骤间等待

按官方价目重估

同等内容若直发 Anthropic 的估算 ⓘ

$0.1826

Anthropic: Claude Sonnet 4.6 (部分)

$0.1594

Anthropic: Claude Sonnet 4.6

$0.2067

Anthropic: Claude Sonnet 4.6

通道指纹

来源步骤 knowledge_recall digit_count digit_count

平台 direct (100%) direct (100%) direct (100%)

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

CDN - - -

ID 格式 msg_0189aDMoW6vojUhbePcvKsry msg_01WBVJCCRGGYxrjrWb2bTzPQ msg_01Gdd3wxoT2wCE7a6UQ5GJct

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-L2

cf-ray: a11a7a804fd3b4c4-LAX
server: Photon-Edge
set-cookie: _cfuvid=vbGvNwJ7HN8EglesNtZG4EVAYP7OVW9CGhacRjRoLGI-1782458731.571253-1.0.1.1-88NgrgI1KJIG7jmH31..ttYnVdv6RlcG8X4E9vh0BSI; HttpOnly; SameSite=None; Secure; Path=/; Domain=api.anthropic.com

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

1782458731187-8vriz0l82v3rv
req_011CcRUPpADfhfTXSCTkxweq
a11a7a804fd3b4c4-LAX

R-kiro

via: 1.1 google
cf-ray: f037776dcb245198f6f05a04630c4e47-IAD
server: nginx

id_format: msg_prefix
via_chain: ["1.1 google"]
extra_fields: ["stop_details"]
has_service_tier: true
has_inference_geo: true
model_has_date_suffix: false
usage_has_cache_fields: true

hdr:viabody:id:msg_prefixbody:usage.service_tierbody:usage.inference_geobody:usage.cache_fieldsbody:extra_fieldsupstream:anthropic-direct-from-body

req_f037776dcb245198f6f05a04630c4e47
f037776dcb245198f6f05a04630c4e47-IAD

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_011CcRUQj8PMEiTEHjN7HJWz

协议指纹评分（机器自动）

总分 / 100 93 100

基线状态已对照基线已对照基线

模型匹配 — 10.0 /10 ×14 10.0 /10 ×14

缓存连续性 — 10.0 /10 ×14 10.0 /10 ×14

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

缓存 TTL 一致 — 10.0 /10 ×15 10.0 /10 ×15

缓存命中比 — 10.0 /10 ×20 10.0 /10 ×20

身份(结构) — 10.0 /10 ×7 10.0 /10 ×7

知识截止 — 10.0 /10 ×7 10.0 /10 ×7

身份(自由) — 10.0 /10 ×7 10.0 /10 ×7

原生 msg-ID — 10.0 /10 ×8 10.0 /10 ×8

Req-ID 透传 — 0.0 /10 ×4 10.0 /10 ×4

stop_reason — 10.0 /10 ×3 10.0 /10 ×3

service_tier — 10.0 /10 ×6 10.0 /10 ×6

inference_geo — 10.0 /10 ×5 10.0 /10 ×5

SDK 一致 — 10.0 /10 ×2 10.0 /10 ×2

系统提示纯净 — 0.0 /10 ×8 10.0 /10 ×8

延迟基线 — 10.0 /10 ×5 10.0 /10 ×5

流式投递 — 10.0 /10 ×5 10.0 /10 ×5

综合结论

与基线相似度执行失败，不评分 93 v3.24.1 基准（参考）

雷达图例

共 21 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 中文遵循 instruction_following_lang
20 档位思考量 tier_thinking_volume_match
21 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 2716 ms 6689 ms 6246 ms

输出速度 - 100.9 tok/s 104.1 tok/s

完成步骤 5/6 成功 · 1 失败 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 ⓘ

$0.1826 (部分)

$0.1594

$0.2067

缓存读取占比 ⓘ 71% (部分) 69% 70%

揭晓

通道

服务商 SSSAiCode ↗

通道O-Max-L2

服务商 ClaudeZZ ↗

通道R-kiro

官方基准

返回列表