PART 1 · INFERENCE · LIVE
Host the decision models
Pools: s1-fast/pro (NVIDIA), s1-lite (small GPUs, incl. Jetson Orin) and s1-cpu (CPUs).
REAX is Bittensor subnet 92: an open inference network for decision models. Anyone with a capable GPU can host the models, validators check every answer, and the network rewards correct, fast serving.
01 / HOSTING IS MINING
A decision model answers one focused question: route, escalate or choose what happens next.
Miners run pinned models; validators check answers and measure speed. Correct, fast miners earn more weight.
Pinned models and runtimes make answers verifiable.
Fresh questions each round; wrong answers or model swaps score zero.
Correct answers first, speed second.
Paused tracks follow published run-out rules.
PART 1 · INFERENCE · LIVE
Pools: s1-fast/pro (NVIDIA), s1-lite (small GPUs, incl. Jetson Orin) and s1-cpu (CPUs).
PART 2 · COMPETITIONS · COMPETITION TRACKS PAUSED
Generalist paused since 30 Oct 16:30 UTC; REAX Engine after its 30 Oct slot. No next crowning is scheduled.
02 / THE FIRST PRODUCT
Decision Models is the customer product: a Jev-compatible, euro-billed API.
Decision Modelss1-fast, s1-pro, s1-vision and REAX 1.3. Peer-to-Peer keys can opt in to miner-answered s1-fast requests, checked against our GPU. EU-tier requests never use the network.LIVE · DECISIONMODELS.IO03 / WHY DECISION MODELS
Agents route requests, escalate cases and choose actions. Focused models provide fast, calibrated answers.
04 / GET STARTED
USE
One Jev-compatible API. Euro billing; no on-chain payment.
Use the APIMINE
Setup guide covers hardware, wallet, registration and node.
Start miningVALIDATE
Recompute answers, measure speed, set weights. Public code.
Validator guideREAX ENGINE
A prefill-only inference engine for decision models: one forward pass, no token generation. MIT License: the code is public and free for anyone to use, including companies. Mining and validating on subnet 92 are always free.
Measured with Perplexity Decider v1.1 27B in FP8 on one RTX PRO 6000 Blackwell GPU, against the model's reference server with the same harness.
Experimental shared-state mode: 6.0× (31.73 to 31.98 vs. 5.24 to 5.28 decisions per second) on Decision-Index-style traffic where requests carry several questions about one state; single-question requests stay at about 3.1×. No measurable change in the Decision Index. It is an approximation: 2.1 % of multi-question top-1 answers differ.
05 / REWARDS
06 / STATUS
OPEN NETWORK · LIVE
EU TIER · BY CONTRACT
Live miner scores: miner leaderboard. Source: github.com/reaxlabs. More on the status page.
07 / ROADMAP
LIVE
Mainnet since 14 Oct. Capacity scoring v2 (15 Oct): pool pay follows verified throughput. Public miner board.
PAUSED
Paused: generalist since 30 Oct 16:30 UTC; REAX Engine after 30 Oct. No next crowning. REAX 1 base; Evaluation v3 when active.
EMISSIONS
Miner emission splits between inference and competitions; see the mining page.
LIVE
Live since 1.2.0rc8. Since 26 Oct 20:08 UTC: up to 2,560 decisions/s (s1-fast), 640/s (s1-pro) per hotkey. 1.2.0rc9.post5 added opt-in multinode. Since 21 Oct (block 9331987, 1.2.0rc10.post1), a hotkey can serve inference, hold one competition entry and serve 2–4 models.
PLANNED
Models that read images.
PLANNED
s1-lite (8 GB GPUs, Jetson Orin) and s1-cpu live; real-device calibration next.
PLANNED
Contracted EU operators for regulated traffic (GDPR, EU AI Act).
Questions? Read the FAQ. REAX is operated by productivity-boost.com, Germany.