Pricing
Buy credits, spend them on forecasts. Each forecast is priced from the models that ran and the size of the request, and every response reports what it cost.
Prepaid
1 credit = 1,000 millicredits (mc).
Per model run
Cost of one model run, in millicredits
rate × series slots × ceil(context ÷ 1,024) × ceil(horizon ÷ 64)| Model | API name | Rate (mc) | One series, ≤1,024 context, ≤64 steps | Context billed up to |
|---|---|---|---|---|
| Chronos-2 | chronos2 | 1,000 | $0.00100 | 8,192 |
| TimesFM 2.5 | timesfm25 | 1,000 | $0.00100 | 16,384 |
| TiRex-2 | tirex2 | 1,000 | $0.00100 | 2,048 |
| Toto 2 (313M) | toto2-313m | 1,000 | $0.00100 | 4,096 |
| FlowState r1 | flowstate-r1 | 3,000 | $0.00300 | 2,048 |
| PatchTST-FM r2 | patchtst-fm-r2 | 3,000 | $0.00300 | 8,192 |
Dollar amounts at the base $10 pack; larger packs include bonus credits.
Questions
Each model run costs its rate × series slots × ceil(context ÷ 1,024) × ceil(horizon ÷ 64) millicredits. One series with up to 1,024 points of history and a horizon up to 64 steps costs exactly the model's rate. Explicit mode runs one model, route mode usually one, ensemble mode every compatible model.
No. Credits are prepaid and only spent by forecasts that run. Credits are reserved when a forecast starts and settled from the models that actually ran.
If unsure, use route mode and Ephemeris picks a model for your data. Use ensemble mode for the best accuracy: it beats every single model in the panel on GIFT-Eval, TIME and fev-bench. Name a model in explicit mode only when you need that model.
Yes. Ephemeris is a remote MCP server at /api/mcp with forecast, list_models, get_balance and get_usage tools, authenticated with the same API key as the REST API.
Rates are public here and in GET /api/v1/models. Every forecast response reports the credits charged and the remaining balance.
Start forecasting