Methodology

Visible assumptions, dated sources.

All calculators run in the browser. Most cost formulas multiply token volume by provider list prices per one million tokens, then scale by calls, users, tasks or time. Specialized calculators add the extra workload components stated on the page.

Pricing snapshot — reviewed August 28, 2026

Provider	Model	Input / 1M	Output / 1M

Rates shown are the currently effective standard text-token rates in the site pricing registry.

Primary sources

Known limitations

List prices are not the whole bill. Long-context tiers, prompt caching, cache writes, batch processing, web/search tools, image/audio tokens, regional processing, fine-tuning, vector infrastructure and negotiated discounts can change actual cost. Token estimators are approximate unless a provider-specific tokenizer is used.

Where a model has a published long-context surcharge, TokenCOGS applies it when the per-request input crosses the provider threshold. Provider tool fees, cache-write/storage fees and negotiated plans can still require separate modeling.

Update rule

TokenCOGS is designed around a live pricing registry. A Cloudflare Worker checks supported official provider pricing sources on a schedule, validates parsed values, and updates the registry only when the source passes sanity checks. If a parser fails or a provider changes its page structure, TokenCOGS keeps the last verified values instead of publishing an unvalidated price. The site also ships with a dated fallback snapshot so calculators remain usable if the live registry is temporarily unavailable.

New model launches still require validation before being added to the comparison set. Automatic synchronization is intended to keep prices for already-supported models current; it does not blindly publish every string discovered on a provider page.