Bittensor subnet ships REAX 1.3, tops HF Decision Index in four weeks
Bittensor's decentralized training ecosystem has delivered again: the "decentralized lab" on Subnet 92 trains decision models through open miner competitions, with the official team integrating the best submissions into REAX 1.3, released under Apache-2.0 for open download on Hugging Face. The model tops the Hugging Face Decision Index 0.4.1 leaderboard with a Full score of 74.0, beating out 274 open replication models; a scale-vs-score scatter plot from Florian Tramèr's team account airesearch12 shows REAX 1.3 (roughly 30B scale) occupying the right end of the Pareto frontier at 70+ points, earning it the nickname "Jev-killer" among those citing it.
Confirmed
- REAX 1.3 is open-sourced on Hugging Face under Apache-2.0 (released by constreborn and ReaxLabs)
- Ranks #1 on the HF Decision Index 0.4.1 with a Full score of 74.0, among 274 open replication models ranked (ReaxLabs)
- Training mechanism: miners on Bittensor Subnet 92 train decision models in open competitions, and the official release integrates the best of them; the subnet has been live for only four weeks (Decision Models)
Why it matters
- constreborn's core claim is that decentralized networks solve hard problems faster than companies, and REAX's rapid rise to the top is seen as an empirical case for this thesis
- The project demonstrates a "competition mining + integrated release" decentralized training paradigm, producing the top open-source decision model on the leaderboard within four weeks, offering a counterpoint to closed corporate training models
2026-11-06 ~ 2026-11-07 · 6 related posts
Primary sources
- REAX 1.3 claims #1 on HF Decision Index with 74.0, built via Bittensor subnet in four weeks — airesearch12 · 2026-11-06
- [source] Bittensor subnet 92 releases REAX 1.3, an open-source decision model blending miner models — const_reborn · 2026-11-06
- [source] Bittensor miners train open-source decision model REAX 1.3 — const_reborn · 2026-11-06
- Open-weight REAX 1.3 tops the Pareto frontier with 70+ score — airesearch12 · 2026-11-06
2 near-duplicate retellings: airesearch12 · markjeffrey