Ping Model

Measure AI model response latency with consistent formatting.

Quick Start

Simple Ping (current model)

bash command:"node {baseDir}/ping-model.js"

Ping Specific Model

bash command:"node {baseDir}/ping-model.js --model minimax"

Compare Multiple Models

bash command:"node {baseDir}/ping-model.js --compare kimi,minimax,deepseek"

Command Reference

Command	Description
/ping	Ping current active model
/ping kimi	Switch to kimi, ping, return
/ping minimax	Switch to minimax, ping, return
/ping deepseek	Switch to deepseek, ping, return
/ping all	Compare all available models

Output Format

Required format - ALWAYS use this exact structure:

🧪 PING {model-name}

📤 Sent:     {HH:MM:SS.mmm}
📥 Received: {HH:MM:SS.mmm}
⏱️  Latency:  {formatted-duration}

🎯 Pong!

Latency Formatting Rules

• < 1 second: Display as XXXms (e.g., 847ms)
• ≥ 1 second, < 60 seconds: Display as X.XXs (e.g., 1.23s)
• ≥ 60 seconds: Display as X.XXmin (e.g., 2.5min)

Examples

Fast response (< 1s):

🧪 PING kimi

📤 Sent:     09:34:15.123
📥 Received: 09:34:15.247
⏱️  Latency:  124ms

🎯 Pong!

Medium response (1-60s):

🧪 PING minimax

📤 Sent:     09:34:15.123
📥 Received: 09:34:16.456
⏱️  Latency:  1.33s

🎯 Pong!

Slow response (> 60s):

🧪 PING gemini

📤 Sent:     09:34:15.123
📥 Received: 09:35:25.456
⏱️  Latency:  1.17min

🎯 Pong!

Cross-Model Testing

When testing a non-active model:

1. Save current model context
2. Switch to target model
3. Execute ping
4. Measure latency
5. Restore original model
6. Display result

Critical: Always return to the original model after testing.

Comparison Mode

bash command:"node {baseDir}/ping-model.js --compare kimi,minimax,deepseek,gpt"

Output format:

══════════════════════════════════════════════════
🧪 MODEL COMPARISON
══════════════════════════════════════════════════

🥇 kimi      124ms
🥈 minimax   1.33s
🥉 deepseek  2.45s
4️⃣  openai    5.67s

🏆 Fastest: kimi (124ms)

Implementation

The ping latency is measured as the time between:

• T1: Message received by the agent
• T2: Response ready to send

This captures the model's internal processing time, not network latency.