Hirondel-7M: zero-shot probabilistic time-series forecasting

Hirondel — from French hirondelle (swallow), the bird said to tell the weather.

Hirondel is a 7M-parameter time-series forecasting model for zero-shot probabilistic forecasting: 9 quantiles from 0.1 to 0.9, any horizon, CPU-friendly. Code and weights under Apache-2.0: https://github.com/tncrd/hirondel.

One of the most accurate small time-series foundation models. At release (October 2026), it was the most accurate on GIFT-Eval, fev-bench and TIME among models under 10M parameters with public weights and a permissive license (official leaderboards, read on 2026-10-23; how this was checked: showcase/README.md).

Results

Benchmark	MASE	CRPS (fev-bench: SQL)
GIFT-Eval (97 tasks, normalised by Seasonal Naive; lower is better)	0.7003	0.4816
fev-bench (100 tasks, skill score in %; higher is better)	32.91	44.90
TIME (98 tasks, normalised by Seasonal Naive; lower is better)	0.6551	0.5520

These numbers are computed by us and not yet listed on the leaderboards. Added to the fev-bench leaderboard, they would rank 7 of 33 (SQL) and 6 of 33 (MASE) among the models the site shows by default (2026-10-23).

Univariate setting: every model forecasts each series alone, without covariates (competitors run by us in forced univariate mode with their public packages; TiRex, TimesFM-2.5, FlowState-r1.1 and Moirai-2.0 univariate by design). Pareto fronts over models with a permissive license, the others shown hollow.

Model	Parameters	License	GIFT-Eval MASE	GIFT-Eval CRPS	TIME MASE	TIME CRPS	fev-bench SQL skill
Toto-2.0-4m	4.1M	Apache-2.0	0.7591	0.5269	0.6940	0.5850	40.04
Timestral-5M	5.2M	Modified MIT (revenue cap)	0.7102	0.4836	0.6571	0.5490	44.36
Hirondel (ours)	7.5M	Apache-2.0	0.7003	0.4816	0.6551	0.5520	44.90
Moirai-2.0-small	11.4M	CC-BY-NC-4.0	0.7281	0.5164	0.7030	0.5885	39.31
FlowState-r1.1	18.5M	Apache-2.0	0.7014	0.4901
Toto-2.0-22m	21.9M	Apache-2.0	0.7226	0.5009	0.6751	0.5654	42.37
Timestral-28M	27.8M	Modified MIT (revenue cap)	0.6817	0.4641	0.6455	0.5408	45.86
TiRex	35.0M	NX-AI Community License	0.7213	0.4921	0.6831	0.5731	42.58
Chronos-2	119.5M	Apache-2.0	0.7037	0.4911	0.6659	0.5584	43.97
TimesFM-2.5	231.3M	Apache-2.0	0.7050	0.4903	0.6686	0.5674	42.20

Industrial data: IND-Eval

Five industrial sources of IND-Eval, none in the training data (top row), then the nine of the paper's aggregate (bottom row). Hirondel: 0.839 / 0.536 (nMASE / nCRPS) on the five. With every model forecasting each channel alone (circles, our runs with the same harness), Hirondel is ahead of Chronos-2, TimesFM-2.5 and TiRex, level with Timestral-5M and Toto-2.0-22m, and behind Timestral-28M. Forecasting the channels jointly with PRONTO's covariates (crosses) moves Chronos-2 from 0.895 to 0.822. Tables per source, protocol and model versions: README, IND-Eval.

Official leaderboards, read on 2026-10-23 (some models use covariates or joint multivariate forecasting there). Sources: showcase/data/ in the code repository.

Usage

import numpy as np
from hirondel import HirondelForecaster

forecaster = HirondelForecaster.from_pretrained("Tncrd/Hirondel-7M")
t = np.arange(1000)
context = np.sin(2 * np.pi * t / 24) + 0.1 * np.random.default_rng(0).normal(size=t.size)
forecast = forecaster.forecast(context, prediction_length=48)
print(forecast.quantiles.shape)   # (1, 48, 9)

Install with pip install "hirondel @ git+https://github.com/tncrd/hirondel@v1.0.0" (* pip install hirondel: PyPI release in preparation). Inference details, tail levels, hardware and numerical parity: see the code repository's README (https://github.com/tncrd/hirondel).

Training data

Real data under licenses that allow this release, plus synthetic data. Every source, its license, its attribution and the changes made are listed in DATA.md; NOTICE gathers the required attributions. No dataset of GIFT-Eval's test split, fev-bench or TIME is in the training data.

Limitations

• Univariate: no covariates, no cross-series information.
• Context: 2 048 points at full resolution plus a coarse reading of up to 8 192 points.
• Tail levels 0.01 / 0.05 / 0.95 / 0.99 are extrapolated, not learned.

License

Apache-2.0 for code and weights. NOTICE must be kept when redistributing.

Citation

@misc{hirondel2026,
  title        = {Hirondel-7M: zero-shot probabilistic time-series forecasting},
  author       = {{The Hirondel Author (Tncrd)}},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/Tncrd/Hirondel-7M}}
}