CPEN RouterOpenAI-compatible routing with live price caps
AI analysis API

How much can Gemini users save?

Start with a $1,000/month Gemini 2.5 Flash input bill. CPEN shows the projected bill, monthly savings, and same-GPQA routes before any table.

Create API keyCompare routes
Live now1 routes available1/3 presets liveUpdated 08:26 AM
Price comparison

Same GPQA, different price.

Start with the bill you already know: $1,000/month on reference-model input becomes a concrete CPEN route, savings percent, and dollar delta before the detailed $/1M table.

Live savings calculation

At $1,000/month current input spend

Default view uses input price per 1M tokens. Output and total modes are estimates for comparing token mixes, not metered billing.

Primary scenario

Monthly data analysis / extraction

Gemini 2.5 Flash → CPEN Middle

Current bill$1,000.00
Projected CPEN$266.67
Monthly savings$733.3373%
Switching examples

Savings scenarios as you scroll

Each model shows how the bill changes when moved to a same-GPQA CPEN preset.

Coding and document summariesClaude Sonnet 4.6 → CPEN Middle
Current bill$1,000.00
Projected CPEN$26.67
97%$973.33saved / month
Frontier reasoning workloadGemini 3.5 Flash → CPEN High
Current bill$1,000.00
Projected CPEN$80.00
92%$920.00saved / month
Chatbot and invoice automationGPT-5.4 mini → CPEN Middle
Current bill$1,000.00
Projected CPEN$106.67
89%$893.33saved / month
Light agent callsGemini 3 Flash → CPEN High
Current bill$1,000.00
Projected CPEN$240.00
76%$760.00saved / month

Up to 97% savings

ModelGPQAInput value scoreInput $/1MCPEN input $/1MInput comparison
GPT-5.4 nanoOpenAI7638.0$0.25$0.0580%
GPT-5.4 miniOpenAI8817.0$0.75$0.0889%
GPT-5.4OpenAI925.56$2.50$0.1295%
Claude Haiku 4.5Anthropic78100$0.10$0.0550%
Claude Sonnet 4.6Anthropic87.54.19$3.00$0.0897%
Claude Opus 4.7Anthropic91.42.75$5.00$0.1298%
Gemini 2.5 FlashGoogle82.837.6$0.30$0.0873%
Gemini 3 FlashGoogle90.426.9$0.50$0.1276%
Gemini 3.5 FlashGoogle92.29.31$1.50$0.1292%
GPT-5.4 nanoOpenAI · GPQA 76
Input $/1M
$0.25
CPEN input $/1M
$0.05
Input comparison
80%
GPT-5.4 miniOpenAI · GPQA 88
Input $/1M
$0.75
CPEN input $/1M
$0.08
Input comparison
89%
GPT-5.4OpenAI · GPQA 92
Input $/1M
$2.50
CPEN input $/1M
$0.12
Input comparison
95%
Claude Haiku 4.5Anthropic · GPQA 78
Input $/1M
$0.10
CPEN input $/1M
$0.05
Input comparison
50%
Claude Sonnet 4.6Anthropic · GPQA 87.5
Input $/1M
$3.00
CPEN input $/1M
$0.08
Input comparison
97%
Claude Opus 4.7Anthropic · GPQA 91.4
Input $/1M
$5.00
CPEN input $/1M
$0.12
Input comparison
98%
Gemini 2.5 FlashGoogle · GPQA 82.8
Input $/1M
$0.30
CPEN input $/1M
$0.08
Input comparison
73%
Gemini 3 FlashGoogle · GPQA 90.4
Input $/1M
$0.50
CPEN input $/1M
$0.12
Input comparison
76%

Input value score excludes success rate, output price, and total token cost. Input comparison uses CPEN preset input prices; final cost depends on token mix.

Presets

Three presets. One API.

Low, Middle, High are convenience presets grouped by GPQA score. Override price, RPM, and strategy per workload.

Low$0.05/1MUp to 82 GPQA

Bulk extraction and classification.

Use cpen/low
Middle$0.08/1M82–90 GPQA

Balanced reasoning for invoices and forms.

Use cpen/middle
High$0.12/1M90+ GPQA

Frontier reasoning at a fraction of the cost.

Use cpen/high
How it works

Cap first. Route second. Reject if over.

Set a USD ceiling per 1M tokens. CPEN picks a live route at a matching GPQA score inside your cap. If none fits, the request is rejected — never silently overspent.

1Set a cap

Declare max USD per 1M tokens before dispatch.

2Route live

CPEN selects a live model at a matching GPQA score inside your cap.

3Reject over budget

No fit? Request fails. No silent escalation.

Savings vs reference models within ±2 GPQA points. Final cost depends on token mix and route availability.

Presets

Three presets. One API.

Low, Middle, High are convenience presets grouped by GPQA score. Override price, RPM, and strategy per workload.

PresetsDocs