NXAI TiRex-2

A recurrent forecasting model that brings another approach to patterns within and across numerical series. Broad pretraining is a background in time-series behavior, not evidence of specialist expertise in any particular asset.

Training background

The model card links the Chronos and LOTSA data collections. The research describes real-world series, synthetic processes and cross-variable training. Separate benchmark-specific checkpoints exclude evaluation datasets for cleaner comparisons.

Available through Pretick

Pretick supports multiple targets and past covariates, with up to 2,048 input rows and 320 future points. Longer consensus requests exclude TiRex-2. Timestamped requests include known clock and weekday features. Arbitrary user-supplied future covariates are not exposed.

{
  "model": "nxai/tirex-2",
  "data": [
    102,
    108,
    111,
    104,
    99,
    106
  ]
}

Use this request at POST https://pretick.ai/v1/timeseries/forecast, or select the same model through MCP. Omit model for consensus. Each successful request costs $0.02.

Using this model for price forecasting

Reading the evidence

Pretick serves the main TiRex-2 checkpoint. Results from decontaminated benchmark variants should not be attributed automatically to this hosted checkpoint.

Sources and version

• Model card and checkpoint variants
• Official implementation
• TiRex-2 paper

Hosted checkpoint: NX-AI/TiRex-2
Revision: 05e5b26db52bfb256f1ae1bdf785589850482de3
License/access: Apache 2.0.

Responses identify the exact model version. These pages describe Pretick's integration; the original model may support additional features.