Pricing
Rates per million tokens, how caching is billed, and what a plan includes.
Token rates
Rates are per million tokens in US dollars and are the same on every plan. A plan changes how much usage is included and what limits apply, never the rate at which a token is billed.
| Model | Context | Max output | Input /M | Cached input /M | Output /M |
|---|---|---|---|---|---|
cog-fast | 256K | 32K | $0.600 | $0.060 | $3.00 |
cog-pro | 256K | 32K | $1.20 | $0.120 | $5.00 |
cog-embed | 32K | n/a | $0.020 | n/a | n/a |
Credits
Usage is stated in credits, where 1 credit is 1 cent, so 100 credits is one dollar. Included credits are topped up at the start of each billing cycle and expire at the end of it. Credits you buy on pay-as-you-go do not expire.
Caching
- Cache read: one tenth of the input rate.
- Cache write: 1.25 times the input rate, charged once.
A prefix is cached when it is repeated at the start of a request. For an agent that resends the same system prompt or the same document on every turn, this is usually the largest single line of savings available, and it requires nothing more than keeping that prefix byte-identical.
Embeddings
cog-embed is billed on input tokens only, at the rate shown above. There is no output charge,
because there is no output text.
What you are not charged for
- Failed requests that never reached a model.
- Streaming responses that a client aborted before any token was generated.
- Comment lines, keep-alives and any other protocol overhead.
Seeing the number before you commit
Send one request of the shape your workload uses, read usage from the response, and multiply. Rates
and context ceilings are on Models and prices; what each plan includes
is on Plans.