Google logo

TimesFM 2.5

TimesFMv2.5Current
byGoogleGoogle(enterprise)
Released September 15, 2025
Context16.4K tokens
Price (In / Out)Free / Free
CategoryLarge Language Model
Max Output1.0K tokens

About TimesFM 2.5

TimesFM (Time Series Foundation Model) is a pretrained decoder-only foundation model by Google Research for time-series forecasting. It uses a patched-decoder attention architecture where patches of time-points are treated as tokens, enabling zero-shot forecasting across domains and temporal granularities from minutes to yearly.

Capabilities

textextended thinking

Input Modalities

time-series

Output Modalities

time-seriesquantiles

Technical Details

API Identifier
google/timesfm-2.5-200m-pytorch
Category
Large Language Model
Context Window
16,384 tokens
Max Output Tokens
1,024 tokens

Tags

time-seriesforecastingfoundation-modelgoogle-researchdecoder-onlyzero-shotopen-sourcepytorchbigqueryquantile

Benchmarks

Performance scores for TimesFM 2.5 across standard benchmarks.

Monash Archivegoogle-research · Sep 2025
85%
ETT Long-horizongoogle-research · Sep 2025
78%

Pricing

Token pricing for TimesFM 2.5 API usage.

Input Tokens

Free

per million tokens

Output Tokens

Free

per million tokens

Pricing Calculator

Input cost$0.00
Output cost$0.00
Estimated monthly cost$0.00

Open-source (Apache 2.0). Free to download and run locally. Enterprise options via BigQuery ML and Vertex AI with their respective pricing.

Competing Models

Same pricing tier — direct alternatives to TimesFM 2.5

Efficiency