1,016
1,016 Real Prompts
Mar 2 – Mar 29, 2026
$11.84
Total Savings
vs all-Claude baseline
77.7%
Savings Rate
routed $3.40 vs actual $15.24
0.294
Avg Complexity
out of 1.0 scale
$15.24
Actual Spend
what was charged (all Claude)
$3.40
Routed Cost
what gateway would charge
Cost & Volume Timeline
Daily requests and cost — actual vs routed (Mar 2–29 2026)
Cumulative savings ($0 → $11.84)
Complexity score distribution (0.0 → 1.0 · 20 buckets)
265
156
39
27
78
240
21
6
1
3
2
0
2
2
1
0
0
1
1
171
0.00.250.500.751.0
Model Routing
Routing distribution (n=1,016)
Haiku
807 prompts (79.4%) $0.0008/req
Sonnet
32 prompts (3.1%) $0.003/req
Claude
177 prompts (17.4%) $0.015/req
Complexity range per model tier
Haiku avg 0.138
Sonnet avg 0.370
Claude avg 0.991
0.00.300.600.851.0
Routed cost split by model tier
Cost efficiency — what you paid vs what you needed to pay
All-Claude (actual)
$15.24
Routed (gateway)
$3.40
$11.84 saved (77.7%)
Request Type Analysis
29.8%
summarization
303 prompts
saved $4.18 (92%)
26.8%
general
272 prompts
saved $3.86 (94%)
22.2%
extraction
226 prompts
saved $0.95 (28%)
12.5%
reasoning
127 prompts
saved $1.69 (88%)
8.7%
generation
88 prompts
saved $1.17 (89%)
Savings by request type
summarization
$4.18 (92%)
Average complexity by type
Key insight: Extraction has the highest avg complexity (0.756) because prompts include full clinical notes. General queries average just 0.052 — almost all routed to Haiku.
Top 20 highest-cost prompts — actual vs what gateway would charge
| # | Prompt | Type | Model | Actual $ | Routed $ | Would Save |
| 1 |
Evaluate this You are a senior product engineer and UX systems designer working |
extraction |
Claude |
$0.0150 |
$0.0150 |
0% |
| 2 |
Stop hook feedback: [Preview Required] Code was edited but no dev server is runn |
general |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 3 |
bt13qbmv0 toolu_01GatfEdqu7E |
extraction |
Sonnet |
$0.0150 |
$0.0030 |
80% |
| 4 |
Analyze this You are a senior full-stack engineer working on SignalCare, a healt |
extraction |
Claude |
$0.0150 |
$0.0150 |
0% |
| 5 |
Stop hook feedback: [Verification Required] Code was edited while a preview serv |
general |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 6 |
Do we have auto policy rules in the platform that moves referral state based on |
general |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 7 |
You are a senior full-stack engineer working on SignalCare, a healthcare workflo |
extraction |
Claude |
$0.0150 |
$0.0150 |
0% |
| 8 |
We currently move cases to a queue that belons to a group. So if anyone working |
general |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 9 |
You are a senior full-stack engineer working on SignalCare, a healthcare workflo |
extraction |
Claude |
$0.0150 |
$0.0150 |
0% |
| 10 |
You are a senior full-stack engineer working on SignalCare, a healthcare workflo |
summarization |
Claude |
$0.0150 |
$0.0150 |
0% |
| 11 |
b75cm5rzb toolu_01GsX5DbMuBH |
summarization |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 12 |
I want to enable or disable the various features we last implemented in the UI. |
general |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 13 |
Are you creating a web page in config to do this? |
general |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 14 |
Yes, I wan to make it possible without a redeploy. Is that doable |
generation |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 15 |
Yes, Add the page under System Config. Clearly explain what each feature does wh |
reasoning |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 16 |
b95gsb569 toolu_01L2dUZ4yaWn |
summarization |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 17 |
Update the product rating matrix you built earlier based on what we have now. Sh |
reasoning |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 18 |
Agreed. Create 10 new referrals using the latest template. Create the referrals |
generation |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 19 |
Can you log in to __https://app.signalcare.ai____ what is app__ |
general |
Haiku |
$0.0150 |
$0.0008 |
95% |
| 20 |
Yes. Just give me your opinion. About, like, what can we do in terms of layering |
general |
Sonnet |
$0.0150 |
$0.0030 |
80% |
Activity Patterns
Requests by hour of day (UTC) — peak 8h–20h highlighted
0h6h12h18h23h
Prompt length distribution
297 prompts (<100 chars) are simple queries routed directly to Haiku.
158 prompts (>5k chars) include full clinical notes — these drive extraction cost.
24-hour activity heatmap — hover for exact counts
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
All 1,016 Prompts
ROI Calculator
$75.00
Current cost (all Claude)
$16.73
Optimized (with routing)
Based on your real 77.7% savings · Haiku $0.0008 · Sonnet $0.003 · Claude $0.015 per request