Methodology

Token counting

GPT/OpenAI models are counted with the exact BPE tokenizer (o200k_base) via the gpt-tokenizer library, loaded in your browser on demand. For all other models we show a clearly labeled estimate based on a tested character ratio (~4 characters per token), because their official tokenizers are not freely available client-side. Estimates are marked “est.” in the results, never presented as exact.

Pricing

Every price links to the provider’s official pricing page and shows a verification date. We never invent prices or limits. Records currently marked as seed values are placeholders pending manual re-verification against the source and must not be treated as authoritative until confirmed.

Cost formula

request_cost = input_tokens/1e6 × input_price
             + cached_input_tokens/1e6 × cached_price
             + output_tokens/1e6 × output_price
monthly_cost = request_cost × requests_per_month

Limitations

• Non-OpenAI token counts are estimates and can differ from real billing.
• Provider pricing changes; a stale verification date means re-check the source.
• Prices exclude batch discounts, volume tiers, taxes and image/audio tokens.