# TW3Cast reaches third on GIFT‑Eval benchmark

Digest AI · Research · published 2026-09-25T04:00:00Z

Canonical: https://digestai.news/story/tw3cast-reaches-third-on-gifteval-benchmark

## Summary

TW3Cast is a time‑series forecasting system that achieved the third position out of 130 entries on the GIFT‑Eval benchmark as of 2026‑09‑14. The system does not use agents or language models; instead it relies on a frozen routing table built once on the training split. The table selects among four modes for each of 97 dataset‑frequency‑horizon configurations: a specialist (a LoRA or full fine‑tune of Chronos‑2, TiRex or Toto), a quantile blend, a blend of base models, or a selection tournament. Each decision is made on a backtest and guarded by three bias‑mitigation mechanisms.

The best base model alone scores a mean MASE rank of 33.8, the tournament alone 38.0, and the full router 19.4. All routing tables, expert indices, and score files are released, allowing others to regenerate the leaderboard numbers with a single script.

## Key points

- TW3Cast ranks third on GIFT‑Eval with a mean MASE rank of 19.4
- It uses a frozen routing table selecting among specialists, blends, and tournaments
- The system relies on lightly fine‑tuned Chronos‑2, TiRex, and Toto models

## Why it matters

The approach shows that carefully constructed routing tables can match or exceed agentic systems on forecasting benchmarks, offering a simpler alternative for practitioners.

## Sources

1. [TW3Cast: A Frozen Router of Lightly Fine-Tuned Foundation Models for Time-Series Forecasting on GIFT-Eval, Selected Entirely on the Training Split](https://arxiv.org/abs/2609.28506) (arXiv cs.AI, 2026-09-25, primary source)

## Cite

Digest AI, "TW3Cast reaches third on GIFT‑Eval benchmark", 25 September 2026, https://digestai.news/story/tw3cast-reaches-third-on-gifteval-benchmark

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/tw3cast-reaches-third-on-gifteval-benchmark.json