Methodology

Transparent sourcing and formulas so you can trust every number on this site.

Live API pricing

Per-token prices come from the public OpenRouter models API. We convert their per-token USD strings into dollars per 1 million tokens. Cache read/write rates are shown when OpenRouter exposes them. Data is fetched server-side and revalidated about once per hour so visitors never hit upstream rate limits.

OpenRouter prices can include a small gateway markup versus buying directly from a provider. For high-volume production, always verify against the vendor's own price sheet.

Blended $/1M (3:1)

Blended cost assumes a typical chat mix of three input tokens for every one output token:

blended = (3 × input$/1M + 1 × output$/1M) ÷ 4

Your real mix may differ (e.g. summarization is input-heavy; code generation is output-heavy). Use the cost calculator with your own token counts for a precise estimate.

Performance benchmarks

Intelligence, coding, math, and agentic indices, plus individual evals (GPQA, HLE, LiveCodeBench, Terminal-Bench, and more), come from the Artificial Analysis free data API when an API key is configured. Speed (tokens/sec) and latency (time to first token / first answer token) are their median API measurements. We match models to OpenRouter IDs by normalized creator + name slugs.

Design Arena Elo scores are taken from OpenRouter's embedded benchmark payload when present. Attribution: artificialanalysis.ai.

Token counting

The token counter runs entirely in your browser using OpenAI's o200k_base and cl100k_base encodings via gpt-tokenizer. Non-OpenAI models use different tokenizers, so their actual billed tokens can differ slightly for the same text.

Agent & cache costing

The agent cost simulator models multi-turn tool loops: sticky context is cache-written on turn 1 and mostly cache-read afterward when OpenRouter publishes cache-read rates. The calculator and model pages also support a cache hit % and an approximate 50% batch discount for OpenAI, Anthropic, and Google.

Value score

The value leaderboard ranks paid models by intelligence × √(tokens/sec) ÷ blended $/1M. Day-over-day rank deltas use the previous price snapshot when present.

Subscription plans

Consumer plan prices on /plans are manually curated from official pricing pages (no public API). The page shows the last verification date. Always confirm on the provider's site before purchasing.

Limitations

• Regional and enterprise discounts are not modeled.
• Batch, reserved capacity, and fine-tuning prices are out of scope.
• Some Artificial Analysis fields require their Pro API and are not shown on the free tier.