Section 1 of 1
Interpreting Performance Profiles
metricsperformancebenchmarkingttft
Every time you run a batch or compare models, PromptForge generates a performance profile. Understanding these metrics is critical for production deployment.
P95 Latency (TTFT)Time To First Token. Crucial for streaming UIs. If this is > 2 seconds, your end-users will think the app is broken.
Tokens/SecondThroughput speed. Important for generating long-form content (reports, code). Gemini often wins this metric.
Cost per 1k ConversionsProjected burn rate. Helps you calculate ROI before deploying a prompt to 100,000 users.
Heuristic Quality ScoreAn automated 1-100 score analyzing how strictly the model adhered to your Negative Constraints.