REAX 1.3, built on Bittensor subnet 92, tops Hugging Face Decision Index
Lab post and Decision Models card: open-weights decision model scores 74.0 Full on the HF Decision Index against 274 models; two miner fine-tunes run in parallel; Apache-2.0 weights on decisionmodels/REAX-1.3.
Why it matters
REAX frames decision-making as a product: choose, score and route under a Jev-compatible API, with open weights from a permissionless miner competition that stood up in about four weeks. Topping a public Decision Index matters less as a chat-LM leaderboard win than as a signal that specialised decision models can be assembled outside big labs. Caveats remain: index methodology, domain coverage and whether enterprise buyers will trust Bittensor-lineage weights. Watch independent replications and whether HF keeps REAX 1.3 at #1 after the next refresh.
Key facts
- #1 on Hugging Face Decision Index (claimed)
- Full score 74.0 vs 274 models
- Built via Bittensor subnet 92 miner competition
- Two parallel miner fine-tunes (uids 18 and 83)
- Apache-2.0 weights: decisionmodels/REAX-1.3