Which one you need
| Model API | Swarm API | |
|---|---|---|
| What you send | A prompt | A task, a size and a budget cap |
| What you get | One answer | One finished result from a team of agents |
| Who runs the tools | You do, in your own code | We do: sandboxes, browsers and inboxes included |
| Billed | Per token | Per agent-minute, up to your cap |
Model API rates
| deepseek-v4.1-flash-uncensored | Per million tokens |
|---|---|
| Input | $0.12 (provisional) |
| Cached input | $0.015 (provisional) |
| Output | $0.525 (provisional) |
Just the model, called from your own code: no swarm, no sandboxes and no agent-minutes. OpenAI-compatible, with no minimum and no subscription. It draws on the same credits balance as swarms.
Swarm API rates
| Model tier | Per agent-minute |
|---|---|
| Standard (DeepSeek V4.1 Flash, standard or uncensored) | $0.10 (provisional) |
| Best available (coming soon) | $0.30 (provisional) |
One rate covers everything a swarm uses, including the model. These prices are provisional and may change before general availability. Heavy proxy bandwidth is covered by a fair-use limit.
Sizes and default caps
| Size | Agents at once | Default cap | Typical run |
|---|---|---|---|
| Small | Up to 5 | $10 | 10–20 min |
| Medium | Up to 20 | $50 | 20–40 min |
| Large | Up to 50 | $150 | 30–60 min |
| Max | Up to 100 | You set it | Hours, to your deadline |
A bigger size means more agents at once, so a run finishes sooner. It costs more only if the work uses it. Unused budget is released when the run ends.
How the cap works
- 01
Estimate first
Every run starts with a plan and a cost range. Ask for the plan only if you want to approve it first.
- 02
Warning at 80%
A live event and a webhook when spend reaches 80% of the cap. Part of the budget is kept back so the result can always be merged.
- 03
Pause at the cap
The run stops new work and hands back a partial result. You are never billed above the cap.
- 04
Resume within 7 days
Add budget and the same agents pick up from their saved notes and files. Paused runs are kept at no charge.
Presets and add-ons
- Security preset: priced like any other swarm, on agent-minutes.
- Container-image scans are a flat price per scan, shown before you start. Scheduled tests, the local CLI and the GitHub Action for preview deploys use the same credits.
Questions
- Do I need the Swarm API to use the model?
- No. The Model API stands on its own. You can call the model directly and never start a swarm.
- Is the model included in the agent-minute price?
- Yes. A swarm's rate covers the model. Only direct calls to the Model API are billed per token.
- Can a run cost more than the cap?
- No. A run pauses when it reaches its cap, and you can resume it with more budget.
- Do I pay while a run is paused?
- No. Paused runs are kept for 7 days at no charge. After that the agents' machines are deleted and only the results remain.
- Are there volume discounts?
- Talk to us about committed spend.
Related
- A swarm of AI agents behind one API callSend one task and a budget; a swarm of AI agents works it in parallel, each with its own VM, browser and inbox. One merged result, never billed above your cap.Read more
- An uncensored LLM API built for security workAn OpenAI-compatible uncensored LLM API for security teams. DeepSeek V4.1 Flash with refusals removed: it writes the exploit, the PoC and the fix. Pay per token.Read more