Glossary

Definitions for the metrics and terms used across AIpricly.

Editorial intelligence estimate

Capability estimate (0–100) authored by the AIpricly editorial team, calibrated against public benchmarks (MMLU, HumanEval, GPQA, MATH) and Artificial Analysis where comparable. Will be replaced with AA partner data once the API key arrives.Editorial estimate · all rows →

Arena Elo

Human-preference Elo score (800–2000) from LMArena — blind A/B votes where real users pick the better response without knowing the model. Measures perceived quality, not just benchmarks.LMArena leaderboard →

P50 Latency (TTFT)

Median first-token latency in seconds across our benchmark runs. Lower is faster.

Throughput

Sustained output speed in tokens per second after the first token arrives. Higher is faster.

Context Window

Maximum combined input + output token length. Larger windows allow longer documents, conversation history, and code files.