Plans
Four self-serve plans and two reserved ones, and which limit each step changes.
Every plan bills tokens at the same rate. What changes as you go up is how much is included, how
many requests can be in flight, how far cog-pro reads, how long metadata is retained, and who
answers when something breaks.
| Plan | Price | Included | Concurrency | Notes |
|---|---|---|---|---|
| Pay as you go | from $10 | credits you buy, no expiry | 4 | cog-pro reads 64K input |
| Builder | $20/month | $21 in credits | 8 | cog-pro 128K, batch at half rate, per-key retention, 30-day audit |
| Pro | $100/month | $114 in credits | 8 | the plan most teams land on |
| Max | $200/month | $238 in credits | 4 reserved, burst to 16 | service credits against the availability target, 180-day audit, shared support, excess at a 15% discount |
| Reserved | from $1,990/month | capacity, not tokens | a whole node | single-tenant, zero retention by contract, reproducibility by hash |
| On-prem | from $4,900 plus support | not applicable | your hardware | for data that cannot leave the building |
Prices exclude tax. Annual billing is twelve months for the price of ten on Builder and Pro.
Reading the ladder
- The step from pay-as-you-go to Builder is about not having to think about it: concurrency,
a longer
cog-prowindow, batch at half rate, and retention controls per key. - Max exists for work that has to keep running, and it says so in what it adds: reserved concurrency, service credits against the availability target, and a longer audit trail.
- Reserved is capacity, and the price is a node rather than a token count. You pay for it whether you use it or not, which is what makes it possible to promise the availability and the zero retention that go with it.
- On-prem is for the case where the requirement is that nothing leaves your network.
Excess usage
When included credits run out, usage continues at the pay-as-you-go rate, discounted on the higher plans, against a hard monthly ceiling you set per key and per account. Auto-recharge is off by default. A streaming response is never cut mid-answer: the answer in flight completes and is billed.
What scales with the plan
| Pay as you go | Builder | Pro | Max | Reserved | |
|---|---|---|---|---|---|
| Rate per token | same | same | same | same | same |
| Included credits | none | $21 | $114 | $238 | capacity |
| Concurrency | 4 | 8 | 8 | 4 reserved, burst 16 | node |
cog-pro input | 64K | 128K | 256K | 256K | 256K |
| Retention control per key | no | yes | yes | yes | by contract |
| Audit trail | 30 days | 30 days | 90 days | 180 days | by contract |
| Support | self-serve | self-serve | self-serve | shared | agreed |
Choosing
If you are evaluating, start on pay-as-you-go: it costs nothing to keep an account open and you pay only for what you send. Move to Builder when a project runs continuously. Move to Max when an outage costs more than the plan. Talk to us about Reserved when the requirement is single tenancy, a contractual retention guarantee, or capacity that is yours rather than shared.