Pricing Methodology

Price Data Source

We fetch live pricing from OpenRouter, which aggregates prices from hundreds of AI providers. Prices are refreshed hourly to reflect current market rates.

Intelligence Scores

Model intelligence scores come from Artificial Analysis, an independent benchmarking organization. Their Intelligence Index combines 10 evaluations covering:

• Agents: GDPval-AA (real-world work tasks), τ²-Bench Telecom (tool use)
• Coding: Terminal-Bench Hard, SciCode
• General: Long-context reasoning, knowledge & hallucination, instruction following
• Scientific Reasoning: GPQA Diamond, Humanity's Last Exam, CritPt (physics)

Scores are on a 0-100 scale where higher is better. This composite score provides a holistic view of model capability.

Speed Data

Speed metrics (tokens/second) are also sourced from Artificial Analysis. Their throughput benchmarks measure actual inference speed across different hardware configurations.

Context Caching

Context caching reduces costs by reusing previously processed context. Pricing varies by provider:

• Anthropic: Cache reads receive a 90% discount, but cache writes have a 25% surcharge. See pricing.
• DeepSeek: Fixed rate of $0.14/1M tokens for cache reads. See pricing.
• OpenAI: Cache reads receive a 50% discount. See pricing.
• Google: Cache reads receive a 50% discount. See pricing.
• Others: Default to 50% discount as a fallback estimate.

Calculator Inputs

Input Ratio

The "Input Ratio" slider estimates what percentage of your prompt is static (cacheable) vs. dynamic. A large document context or system prompt is typically static, while the user's question is dynamic.

Tokens/Day

Monthly costs are extrapolated from your daily token estimate assuming consistent usage over 30 days. Real-world usage often varies, so treat this as a baseline projection.