Model panel
Six forecasting foundation models, one API key.
Every model is open-weights and zero-shot: send history, get quantile forecasts, no training. Name a model in explicit mode, let route pick one for your data, or use ensemble, an accuracy-weighted blend of the panel that beats every member on GIFT-Eval, TIME and fev-bench.
| Model | API name | Publisher | Size | Licence | Multivariate | Covariates | Context |
|---|---|---|---|---|---|---|---|
| Chronos-2 | chronos2 | Amazon | 120M | Apache-2.0 | Yes | Yes | 8,192 |
| TimesFM 2.5 | timesfm25 | Google Research | 200M | Apache-2.0 | No | No | 16,384 |
| Toto 2 (313M) | toto2-313m | Datadog | 313M | Apache-2.0 | Yes | No | 4,096 |
| TiRex-2 | tirex2 | NXAI | 38M univariate, 82M multivariate | Apache-2.0 | Yes | Yes | 2,048 |
| PatchTST-FM r2 | patchtst-fm-r2 | IBM Granite | 385M | OpenMDW-1.0 | Yes | No | 8,192 |
| FlowState r1 | flowstate-r1 | IBM Granite | 9M | Apache-2.0 | No | No | 2,048 |
Live health, horizon limits, ensemble weights and prices: GET /api/v1/models, or the list_models tool over MCP.
Use them from an agent