Models & pricing

tokcost never asks you for an encoding directly — you pass a model name and it resolves the encoding for you. Whatever encoding it lands on is printed next to every result, so the number is always traceable.

Model → encoding mapping

The match is prefix-based and case-insensitive, so versioned names like gpt-4o-2024-08-06 or claude-3.5-sonnet-latest resolve to the same encoding as their base name.

Model names	Encoding	Exact?
`gpt-4o`, `gpt-4.1`, `gpt-5`, the `o`-series (`o1` / `o3` / `o4`), `chatgpt-4o`	`o200k_base`	Yes
`gpt-4` and `gpt-4-*`, `gpt-3.5`	`cl100k_base`	Yes
`claude-*`	`o200k_base`	Approximate
`gemini-*`	`o200k_base`	Approximate
anything unrecognized	`o200k_base`	Best-effort

The older OpenAI families (gpt-4, gpt-3.5) are checked before the broad gpt-4o / gpt-4.1 prefixes, so gpt-4 lands on cl100k_base while gpt-4o lands on o200k_base. Anything tokcost doesn’t recognize falls back to the modern o200k_base encoding rather than failing.

The approximation caveat

Token counts are exact for OpenAI models — gpt-tokenizer ships the same BPE merges OpenAI uses.

Anthropic and Google do not publish a JavaScript tokenizer. For claude-* and gemini-* model names, tokcost approximates with OpenAI’s o200k_base encoding. The counts will be close but are not ground truth — treat them as estimates for budgeting and gut-checks, not as the exact number those providers will bill you for.

The default model is gpt-4o (o200k_base) when you don’t pass -m.

Pricing table

The --cost flag multiplies your token count by an approximate input price (USD per 1,000,000 tokens) from a small, hand-maintained table:

Model	Input price (USD / 1M tokens)
`gpt-4o`	$2.50
`gpt-4o-mini`	$0.15
`gpt-4.1`	$2.00
`o3`	$2.00
`claude-3.5-sonnet`	$3.00
`claude-3.5-haiku`	$0.80
`gemini-2.0-flash`	$0.10

Price lookup tries an exact match first, then a normalized/prefix match — so a versioned name like claude-3.5-sonnet-20241022 still resolves to the claude-3.5-sonnet row.

Pricing caveats

Input only. The table prices input tokens. Output tokens, cached tokens, and other line items are not modeled.
Approximate, not billing. Providers change prices often. These figures are a rough gut-check — never reconcile a bill against them.
Unknown models report unavailable. If a model isn’t in the table, the CLI prints cost unavailable (and the JSON cost.available is false).

Editing the table

The prices live in src/models.ts as the INPUT_PRICE_PER_MTOK record. Edit that table to add models or update figures to match your contract, then rebuild. The same table backs both the CLI’s --cost flag and the library’s estimateCost.