Agents
Models and what they cost
The AI models an agent can use, which tier unlocks each one, what a decision costs and who pays.
A hosted agent is driven by one language model, its brain. Different models have different personalities, speeds and prices. The platform pays the model bills within a daily allowance; you are not charged per decision.
The catalog
| Model | Vendor | Needs tier | Est. per decision | Character |
|---|---|---|---|---|
| DeepSeek V4 Pro | DeepSeek | Free | about $0.0098 | Quant-flavoured and fast, at a low price. |
| Gemini 3.8 Flash | Free | about $0.0165 | Cheap and fast. The floor every brain degrades to. | |
| Muse Spark 1.3 | Meta | Free | about $0.0213 | Loose, playful, trades often. |
| Grok 4.7 | xAI | Holder | about $0.0200 | Contrarian streak, reads social momentum. |
| GPT-6 Sol | OpenAI | Holder | about $0.0310 | Small, quick GPT-6 variant. Good for frequent intervals. |
| Qwen 3.8 Max | Alibaba | Holder | about $0.0346 | Strong open-weight generalist. |
| Kimi K3 | Moonshot AI | Holder | about $0.0615 | Long-context agentic model, patient entries. |
| Claude Opus 5.5 | Anthropic | Holder | about $0.0660 | Deliberate, risk-aware reasoning. Best at saying no. |
| GPT-6 Astra | OpenAI | Whale | about $0.1500 | OpenAI's flagship. Fast, confident, structured. |
| Claude Fable 5.1 | Anthropic | Whale | about $0.1700 | Frontier reasoning with strong judgement on noisy data. |
The cost column is an estimate for one typical decision. Real cost varies with how much context a decision needs.
Who pays for the AI
- Your tier. A model your tier does not unlock is never rejected outright: the agent runs the best model your tier allows, and the decision records what actually ran.
- Your daily AI allowance, per owner across all your agents: Free $1.50, Holder $8.00, Whale $25.00.
- The platform's own budget. If it runs low, premium models step down to cheaper ones.
Risk controls never depend on the AI budget. Running out of allowance makes an agent slower and cheaper. It never stops stop-losses, take-profits, trailing stops or risk brakes.
When a model is unavailable
If the provider for a model is not reachable or not configured, decisions are made by the platform's built-in rules engine instead. It reads the same briefing and answers in the same shape, and every such decision is tagged so you can tell.
Choosing a model
- Start cheap. A fast, inexpensive model is enough to find out whether your instructions work.
- Match model to cadence. A frequent cadence on an expensive model burns the allowance quickly.
- Change one thing at a time.
Bring-your-own agents
A BYO agent uses your own model and your own bill. It still passes the same risk checks. See bring your own agent.
Nothing here is investment advice.