Nixtla Pricing Overview

I was initially skeptical about replacing our tree-based models with TimeGPT, but the improvements have been undeniable. Not only has the forecast accuracy significantly increased, but the speed and scalability have also exceeded our expectations. TimeGPT has completely transformed the way we handle stock predictions, and we're now delivering more reliable forecasts faster than ever before. Review collected by and hosted on G2.com.

Nixtla is a fantastic company with fantastic products. They have really helped us a lot. There is nothing negative that I can honestly say about them! Review collected by and hosted on G2.com.

From Nixtla’s ecosystem, I particularly value TimeGPT, which I used in a project focused on predicting kNDVI (Normalized Difference Vegetation Index) across 214,351 time series derived from a geospatial datacube. The model demonstrated adaptability; it effectively captured vegetation dynamics when provided with sufficient historical context.

With fine-tuning and the inclusion of exogenous variables, TimeGPT achieved reliable long-horizon forecasts, outperforming traditional models across multiple metrics while better reproducing the shape, amplitude, and temporal dynamics of the kNDVI signal.

TimeGPT also shows strong potential for gap-filling and temporal interpolation in satellite-derived time series, where missing observations are frequent due to cloud cover or sensor limitations. When given full historical context, it generalizes the underlying temporal patterns remarkably well, making it a valuable tool for Earth observation and remote sensing applications, provided adequate computational resources and contextual information are available. Review collected by and hosted on G2.com.

While TimeGPT is a powerful and well-designed model, adapting it to large-scale geospatial forecasting presented several challenges. The main limitation lies in its lack of native support for geospatial data structures, such as data cubes or spatial indexing systems (e.g., H3, S2). To forecast kNDVI signals, each grid cell had to be manually converted into a string-based unique identifier combining latitude and longitude, and the forecasts had to be executed in multiple batches due to API limits.

TimeGPT also relies on inferred temporal frequencies and predefined exogenous variables, but does not yet provide feature importance or interpretability diagnostics, making it difficult to identify which variables drive the forecasts.

Despite these challenges, these limitations are understandable given the model’s original design for univariate time series and represent opportunities for future development — particularly toward native geospatial compatibility and improved model interpretability. Review collected by and hosted on G2.com.