Methodology

Data Sources

FindLLM aggregates data from three sources: Artificial Analysis (benchmarks, quality scores, speed metrics), OpenRouter (pricing, context lengths, provider availability), and HuggingFace (downloads, licenses, open source metadata).

Quality Index

The Quality Index (also called Intelligence Index) is a composite score developed by Artificial Analysis that combines multiple benchmark results into a single comparable metric. It considers performance across reasoning, knowledge, coding, math, and instruction following tasks.

Benchmarks

We track the following benchmarks:

• MMLU-Pro — Massive Multitask Language Understanding (enhanced)
• GPQA Diamond — Graduate-level science questions
• LiveCodeBench — Real-time coding benchmark
• AIME 2025 — American Invitational Mathematics Examination
• MATH-500 — Mathematical problem solving
• SciCode — Scientific coding tasks
• IFBench — Instruction following benchmark
• LCR — Long Context Recall
• TerminalBench Hard — Terminal/CLI tasks
• HLE — Humanity's Last Exam

Speed Metrics

Output Speed is measured in tokens per second — how fast a model generates output. TTFT (Time to First Token) measures latency — how quickly the model starts responding. Both are measured using standardized 1,000-token prompts via the Artificial Analysis testing infrastructure.

Pricing

Prices are shown in USD per 1 million tokens. The blended price assumes a 3:1 output-to-input ratio, reflecting typical conversational usage patterns. Input and output prices are also available separately.

Update Frequency

HuggingFace data syncs hourly, OpenRouter pricing every 4 hours, and Artificial Analysis benchmarks daily. This ensures comparisons reflect the latest changes across all sources.

Glossary of Terms

Quality Index

Composite score combining multiple benchmark results into a single comparable metric.

Blended Price

Cost per 1M tokens assuming a 3:1 output-to-input ratio, reflecting typical conversational usage.

Output Speed

Tokens per second — how fast a model generates output.

TTFT

Time to First Token — latency before the model starts responding.

Benchmark

Standardized test measuring model performance on specific tasks.