Google Research · hosted on Ephemeris
TimesFM 2.5 API
Google's decoder-only forecaster with the longest context in the panel, suited to long histories with slow seasonality.
Explicit mode
"model": "timesfm25"API nametimesfm25
PublisherGoogle Research
Weightsgoogle/timesfm-2.5-200m-pytorch
Size200M
LicenceApache-2.0
MultivariateNo
CovariatesNo
Context read16,384 most recent points
01
REST
Call TimesFM 2.5 directly
cURLUTF-8
curl --request POST \
--url https://ephemeris.cascade.industries/api/v1/forecast \
--header "Authorization: Bearer pc_live_your_key" \
--header "Content-Type: application/json" \
--data '{
"mode": "explicit",
"model": "timesfm25",
"series": [{ "values": [112, 118, 132, 129, 121, 135, 148, 148], "freq": "M" }],
"horizon": 12,
"quantiles": [0.1, 0.5, 0.9]
}'02
Agents
From Claude, Cursor or any MCP client
Add the Ephemeris MCP server, then ask your agent to forecast with timesfm25. The forecast tool takes mode: "explicit", model: "timesfm25".
03
Accuracy
How it scores
On GIFT-Eval (97 configurations, official harness), TimesFM 2.5 alone scores CRPS 0.4923 and MASE 0.7096 as a ratio to seasonal naive (lower is better). The Ephemeris ensemble, which includes it, scores 0.4662 and 0.6841. Full benchmark results.
No GPU to run