AI credits rate card
This is Driftstack's published rate card for AI credits, the one the Terms of Service (section 8.11) refer to. It lists what each model costs in AI credits when the AI agent runs on Driftstack's model access. How many AI credits your plan includes each month is on the pricing page.
How use is counted
- One AI credit is worth US$0.01.
- Each message the AI agent works on calls a model one or more times. Each call uses the tokens the model counts, and each kind of token is charged at its own rate on the card: input, output, cache reads and cache writes.
- A call that is stopped or cut off before it ends is charged the input and cache tokens counted and its output at the most that call was allowed to write. A call that was sent but returned no count is charged the most that call could cost.
- Rates are in AI credits per million tokens. For example, on Claude Sonnet 5, a call that reads 20,000 input tokens and writes 1,000 output tokens uses 10 AI credits (input at 400 and output at 2,000 per million tokens).
- Each model's row also sets the most AI credits a single message can use ("Most one message can use").
- Use on your own model provider key is billed by your provider and uses no AI credits.
Version 1
In force until version 2 takes effect on 18 October 2026 (UTC).
| Model | Input | Output | Cache read | Cache write (5 minutes) | Cache write (1 hour) | Most one message can use |
|---|---|---|---|---|---|---|
Claude Sonnet 5 claude-sonnet-5 | 400 | 2,000 | 40 | 500 | 800 | 60 |
Claude Sonnet 4.6 claude-sonnet-4-6 | 600 | 3,000 | 60 | 750 | 1,200 | 90 |
Claude Haiku 4.5 claude-haiku-4-5 | 200 | 1,000 | 20 | 250 | 400 | 30 |
Rates in AI credits per million tokens; the last column in AI credits. Cache writes are charged at the rate for how long the prompt is cached (5 minutes or 1 hour).
Version 2
Announced ahead of time. Takes effect on 18 October 2026 (UTC). From that date it replaces version 1.
Version 2 adds Claude Sonnet 5.5 at the same rates and limit as Claude Sonnet 5. Every other row is the same as on version 1. It does not move any session to another model: a session that names a model keeps that model, at the same rates. Only a session that names no model runs on the default model, which the create request names.
| Model | Input | Output | Cache read | Cache write (5 minutes) | Cache write (1 hour) | Most one message can use |
|---|---|---|---|---|---|---|
Claude Sonnet 5.5 claude-sonnet-5-5 | 400 | 2,000 | 40 | 500 | 800 | 60 |
Claude Sonnet 5 claude-sonnet-5 | 400 | 2,000 | 40 | 500 | 800 | 60 |
Claude Sonnet 4.6 claude-sonnet-4-6 | 600 | 3,000 | 60 | 750 | 1,200 | 90 |
Claude Haiku 4.5 claude-haiku-4-5 | 200 | 1,000 | 20 | 250 | 400 | 30 |
Rates in AI credits per million tokens; the last column in AI credits. Cache writes are charged at the rate for how long the prompt is cached (5 minutes or 1 hour).
Models not on AI credits
These models run only on your own model provider key. They do not run on AI credits.
- Claude Fable 5.1
claude-fable-5-1 - Claude Opus 5.5
claude-opus-5-5 - Claude Opus 5
claude-opus-5 - Claude Opus 4.8
claude-opus-4-8 - Claude Opus 4.7
claude-opus-4-7
These models are not on AI credits either. You can run them on your own key.
- Claude Haiku 5.5
claude-haiku-5-5
Changes to the rate card
As the Terms of Service (section 8.11) say, Driftstack publishes every change to the rate card and records it in the changelog.
- After Launch, a change that raises the rate of a model by more than 10%, compared with that model's rate 30 days earlier, or that removes a model from the rate card, takes effect at least 30 days after Driftstack announces it; before Launch it takes effect when it is published.
- Any other change, including adding a model, replacing a model, lowering a rate or raising a rate by 10% or less, takes effect when it is published.
For a question about the rate card, email [email protected].