LLM Reference

TimesFM Models by Google

GoogleNoncommercialOpen weights
1 model2026

Last refreshed 2026-09-01. Next refresh: weekly.

Details

ResearcherGoogle
Commercial useCommercial use: non-commercial
Models1
Released2026

About

TimesFM is Google Research's family of decoder-only time-series foundation models for zero-shot forecasting. Earlier checkpoints (through TimesFM-2.5) were univariate; TimesFM-3 adds native multivariate forecasting with past and past-future covariates. Public checkpoints are on Hugging Face; GitHub code is Apache-2.0. BigQuery AI.FORECAST currently documents TimesFM-2.5; TimesFM-3 BigQuery integration is described as coming, not live.

Decision facts

Best fit
forecastingcodingagent workflows
Capability starting point
TimesFM-3
Lowest tracked input
Not tracked
Closest related family
Google Cloud Text-to-Speech

Current Variants

Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.

1 in view
TimesFM-3Current

Use when the workload needs forecasting and 330M parameters.

2026-08forecasting330M parameters

Release Timeline

1 release group
2026-08
1 current
TimesFM-3
forecasting330M parameters
Current

Specifications(1 models)

TimesFM model specifications comparison
ModelReleasedParameters
TimesFM-32026-08330M