Observability / Analytics

Analytics

Data from Jun 02 to Jun 03
Requests
128.4K
▲ 12.3%
Tokens In
42.1M
▲ 8.1%
Tokens Out
8.3M
▲ 5.4%
Cost (USD)
$284.50
▲ 4.2%
Latency P95
2.84s
▼ 9.2%
Rescued ✦
847
▲ fallback OK

Request Rate by Model

claude-opus-4-7 gpt-4o deepseek-v3 qwen-max
00:00 04:00 08:00 12:00 16:00 20:00 23:59

Top Models

ModelReqCost
claude-opus-4-754.2K$152.30
gpt-4o38.1K$78.40
deepseek-v321.5K$22.10
qwen-max9.8K$18.60
claude-haiku-4-54.7K$13.10

Usage — Model × Time

< 70% 70-90% > 90%
Model Minute Hour Day Month Year
All models
R 8.5KW 3.5K
12K / 50K24%
R 240KW 100K
340K / 1M34%
R 4MW 1.8M
5.8M / 8M72%
R 29MW 13M
42M / 100M42%
R 340MW 146M
486M / 1.2B40%
claude-opus-4-7
R 6.4KW 2.8K
9.2K / 10K92%
R 130KW 56K
186K / 240K78%
R 2MW 800K
2.8M / 3.5M80%
R 15MW 6M
21M / 60M35%
R 170MW 70M
240M / 720M33%
gpt-4o
R 1.5KW 600
2.1K / 20K10%
R 56KW 24K
80K / 400K20%
R 1.1MW 500K
1.6M / 4M40%
R 10MW 4M
14M / 40M35%
R 120MW 48M
168M / 480M35%
deepseek-v3
R 360W 150
510 / 10K5%
R 30KW 12K
42K / 200K21%
R 700KW 280K
980K / 2M49%
R 3.7MW 1.5M
5.2M / 20M26%
R 41MW 17M
58M / 240M24%
qwen-max
R 85W 35
120 / 10K1%
R 8.5KW 3.5K
12K / 200K6%
R 295KW 125K
420K / 2M21%
R 1.5MW 600K
2.1M / 20M10%
R 14MW 6M
20M / 240M8%
claude-haiku-4-5
R 3.2KW 1.4K
4.6K / 20K23%
R 95KW 40K
135K / 400K34%
R 1.8MW 800K
2.6M / 4M65%
R 12MW 5.5M
17.5M / 40M44%
R 140MW 60M
200M / 480M42%
Model Minute Hour Day Month Year
All models
R-cache4.2K
C-cache1.1K
hit ratio 79/21
R-cache180K
C-cache42K
hit ratio 81/19
R-cache3.1M
C-cache720K
hit ratio 81/19
R-cache22M
C-cache5.4M
hit ratio 80/20
R-cache260M
C-cache62M
hit ratio 81/19
claude-opus-4-7
R-cache3.4K
C-cache820
hit ratio 81/19
R-cache135K
C-cache31K
hit ratio 81/19
R-cache2.2M
C-cache530K
hit ratio 81/19
R-cache15.5M
C-cache3.7M
hit ratio 81/19
R-cache185M
C-cache44M
hit ratio 81/19
gpt-4o
R-cache620
C-cache180
hit ratio 78/22
R-cache32K
C-cache8.4K
hit ratio 79/21
R-cache680K
C-cache142K
hit ratio 83/17
R-cache5.1M
C-cache1.2M
hit ratio 80/20
R-cache58M
C-cache13.5M
hit ratio 81/19
deepseek-v3
R-cache140
C-cache65
hit ratio 68/32
R-cache9.5K
C-cache2.1K
hit ratio 82/18
R-cache185K
C-cache38K
hit ratio 83/17
R-cache1.1M
C-cache240K
hit ratio 82/18
R-cache12.5M
C-cache2.8M
hit ratio 82/18
qwen-max
R-cache30
C-cache8
hit ratio 79/21
R-cache2.4K
C-cache480
hit ratio 83/17
R-cache52K
C-cache9.8K
hit ratio 84/16
R-cache340K
C-cache65K
hit ratio 84/16
R-cache3.8M
C-cache720K
hit ratio 84/16
claude-haiku-4-5
R-cache1.8K
C-cache420
hit ratio 81/19
R-cache65K
C-cache14.5K
hit ratio 82/18
R-cache1.2M
C-cache280K
hit ratio 81/19
R-cache8.4M
C-cache1.9M
hit ratio 82/18
R-cache96M
C-cache22M
hit ratio 81/19
Active users
847
▲ 6.1%
New users
42
▲ 31%
Power users (>10k tok/d)
68
▲ 8.3%
Idle (no req 14d)
123
▼ 2.1%

Top users — by token

USERWORKSPACETEAMREQ 30DTOKENS INTOKENS OUTCOSTTOP MODEL
amy.chenprojectx-prodProjectX Core48.2K3.1M340K$182.40claude-opus-4-7
peter.zhaoprojectx-prodProjectX Core31.5K2.4M280K$141.20claude-opus-4-7
jackie.linaigw-platformPlatform Eng28.9K1.9M220K$98.30gpt-4o
linda.wangmydeck-prodMyDeck22.1K1.4M180K$72.50gpt-4o
tony.wongprojectx-prodProjectX QA19.8K1.2M140K$64.10deepseek-v3
sarah.lioa-prodOA Core15.3K980K110K$48.80claude-haiku-4-5
kevin.zhouaigw-platformPlatform Eng12.7K820K95K$41.20gpt-4o
mike.chenprojectx-prodProjectX QA10.4K680K75K$32.60qwen-max

Distribution — by workspace

WORKSPACEUSERSACTIVE 30DREQUESTSTOKENSSHARE
projectx-prod32429854.2K18.4M42%
aigw-platform18617231.5K9.8M24%
mydeck-prod14212822.8K7.2M18%
oa-prod988612.4K4.1M10%
projectx-staging52415.8K1.8M4%
oa-staging45221.7K800K2%
Total errors 24h
412
▼ 14%
Error rate
0.32%
▼ 0.08pp
4xx (client)
298
▼ 18%
5xx (upstream)
114
▲ 4%

By status code

CODEMEANINGCOUNT 24HSHARETOP MODELTREND 24H
429Rate limit / quota exceeded14234%claude-opus-4-7▁▂▃▅▇▆▄▃
401Invalid API key8621%—▂▃▂▃▂▃▂▃
503Upstream unavailable6817%gpt-4o▁▁▂▃▄▆▅▃
400Bad request (schema)5213%claude-opus-4-7▃▂▃▂▃▂▃▂
502Upstream gateway error287%qwen-max▁▂▁▂▃▂▁▂
504Upstream timeout184%deepseek-v3▁▁▂▁▂▁▂▁
413Payload too large123%claude-opus-4-7▁▁▁▂▁▁▁▂
othersMisc61%—▁▁▁▁▁▁▁▁

Recent errors

TIMECODEMODELUSERWORKSPACEMESSAGE
12:42:18429claude-opus-4-7amy.chenprojectx-prodRate limit exceeded for tenant minute window (10K)→
12:41:55503gpt-4ojackie.linaigw-platformupstream connection reset by peer→
12:41:03401———Invalid API key (prefix sk-aigw-x…f3)→
12:40:48400claude-opus-4-7linda.wangmydeck-prodmessages[3].content: Expected string, got null→
12:39:21429claude-opus-4-7peter.zhaoprojectx-prodTeam minute quota exceeded (8K/8K)→
12:38:14504deepseek-v3kevin.zhouaigw-platformupstream timeout after 60s→
Hit ratio 24h
81.4%
▲ 2.3pp
Cached tokens
14.2M
▲ 18%
Estimated savings
$184.20
▲ 22%
Avg TTL
8.4 min
±0

By model — last 24h

MODELREQUESTSCACHE READSCACHE WRITESHIT RATIOTOKENS SAVEDSAVINGS $
claude-opus-4-754.2K128K28K82%8.4M$112.30
gpt-4o38.1K82K18K82%3.2M$38.40
deepseek-v321.5K38K9K81%1.4M$14.20
qwen-max9.8K12K3K80%680K$10.80
claude-haiku-4-54.7K8.4K2K81%520K$8.50

Top cached prompts

PROMPT PREFIXMODELHITS 24HTOKENS / HITSAVINGS
You are an expert code reviewer. Analyze the following…claude-opus-4-78,4202,140$42.80
System: You are ProjectX, a workplace assistant for…claude-opus-4-76,1801,890$28.40
[INST] You are MyDeck assistant. Context: user is…gpt-4o4,8201,640$18.20
Analyze the following pull request diff and return…deepseek-v33,1402,820$8.60
Total feedback 30D
2,847
▲ 12%
👍 Positive
2,412
84.7%
👎 Negative
435
15.3%
Avg rating
4.31
▲ 0.08

By model

MODELFEEDBACK👍👎RATIOAVG RATING
claude-opus-4-71,4201,27814290%4.62
gpt-4o82068413683%4.18
deepseek-v33402687279%4.02
qwen-max1821285470%3.74
claude-haiku-4-585543164%3.51

Recent comments

TIMEUSERMODELCOMMENT
12:38👍amy.chenclaude-opus-4-7"完美，一次过。代码 review 提的 3 个问题全是真问题。"
12:22👎linda.wanggpt-4o"reasoning 跳步，要我手动 verify 才发现算错了"
11:48👍jackie.linclaude-opus-4-7"Portkey config 改对了，符合预期"
11:30👎peter.zhaoqwen-max"中文输出夹英文，prompt 没遵守"
10:55👍tony.wongdeepseek-v3"成本省一半，质量基本一致"
10:18👍sarah.liclaude-haiku-4-5"分类任务足够用，比 Opus 快 4x"

Monthly Summary — May 2026

Generated Jun 01 03:00 UTC · scope Accenture China tenant
Total requests
3.84M
▲ 18% vs Apr
Total tokens
1.42B
▲ 22% vs Apr
Total cost
$8,540
▲ 16% vs Apr
Cache savings
$1,820
▲ 28% vs Apr
Avg latency p95
2.91s
▼ 4% vs Apr
Error rate
0.38%
▼ 0.12pp vs Apr

By workspace — May 2026

WORKSPACEREQUESTSTOKENSCOST% OF TOTALΔ vs APR
projectx-prod1.61M596M$3,58642%▲ 21%
aigw-platform922K341M$2,05024%▲ 14%
mydeck-prod691K256M$1,53718%▲ 18%
oa-prod384K142M$85410%▲ 8%
projectx-staging154K57M$3424%▼ 6%
oa-staging77K28M$1712%▲ 3%

Highlights

  • ProjectX 增长最猛：requests ▲ 21%，主要来自 ProjectX Core 团队上线 PR-review agent。
  • Cache 命中率创新高：从 78.2% 升至 81.4%，节省 $1,820（▲ 28%），主因 system prompt 标准化。
  • 错误率持续下降：0.50% → 0.38%，5xx 大量减少（gpt-4o 上游稳定性改善）。
  • 新模型采纳：claude-opus-4-7 占总请求 42%（4 月 28%），claude-3-opus 下线完成。
  • 预警：projectx-prod minute quota 在 09:00-11:00 多次触顶 92%，建议下月扩容至 12K。