How TokenCalc AI sources, reviews, and publishes LLM pricing, specs, and benchmark data — and where the numbers can be wrong.
Prices come from the official provider pricing pages (OpenAI, Anthropic, Google, DeepSeek, Meta, Mistral, xAI, Alibaba, Cohere). Each entry records a pricingSourceUrl and a pricingCheckedAt date. We do not scrape or auto-update; updates are manual and dated.
Intelligence scores, throughput, and latency come from the Artificial Analysis API where available. Entries without a benchmark are shown as "Unrated" rather than estimated. Each entry records a performanceSourceUrl and performanceCheckedAt.
The catalog is reviewed in dated batches (see the changelog). Retired models are kept for archive only, marked noindex, follow, and point to a successor where one exists. Interactive tools and rankings use only the active set.
Found a stale price or a missing model? Contact us and we'll review it in the next batch.