Developer documentation
Build on REAX
REAX is Bittensor subnet 92 (SN92), live on mainnet since 14 October 2026. Miners serve fast, calibrated, finite-answer decision models that are Jev-compatible, and validators measure whether those models are served faithfully. Registration is open, and the source is public.
Status · 21 October 2026
REAX is live on Bittensor mainnet as netuid 92 (since 14 October 2026). Registration is open; the chain sets the burn cost, so check it before you register. Registration is a burn; the cost is set by the chain, so check it on chain or on taostats before you register. There is no public REAX testnet; practise on a local chain with the miner kit. Release 1.2 has four pools: s1-fast, s1-pro, s1-lite and s1-cpu. Our validator (UID 3) runs rounds about every 2 minutes. The source is public at github.com/reaxlabs/reax-subnet, with reax-miner-kit and reax-engine. To start mining, point your coding agent at mine.decisionmodels.io or follow Mine REAX in 30 minutes.
Release 1.2: the current mechanism
The current mechanism (release 1.2) is inference mining in four pools: Inference mining, Verification & calibration, and Competition mining. The generalist and REAX Engine tracks are paused after their Friday 30 October slots; no next crowning is scheduled. See current status on the competition page. Pages written for spec revision 0.6 describe the earlier design.
What REAX does
A REAX request carries a question with a finite candidate set: yes or no, one option out of several, or a level on an ordered scale. A miner answers with a complete probability distribution over that set, signed with its serving key. The network is built for decisions that must come back quickly and with calibrated probabilities, not for free-form text generation.
- Miners serve approved models only. The v1 model catalog is closed: each public
modelalias maps to a signed list of approved model manifests, and miners compete on faithful, available, verified-capacity serving of those models. They do not bring their own models. - Validators measure faithful serving: identity and quality fidelity to the reference manifest, availability, latency (capped at the reference) and verified serving capacity. These sit behind hard eligibility gates for tier, identity and integrity. Validators convert the result into miner weights.
- Chain emissions are an incentive mechanism. They are not customer billing and not a service guarantee. Customer charges and any operator settlement are fiat-accounted and kept separate from the score ledger.
Architecture
Customer requests never reach a miner directly. They enter through the Decision Models API, are authenticated and projected by the REAX gateway, and are dispatched to an eligible miner in the right tier. Validators inject their probes through the same gateway, so a probe travels the same path as any other request.
Every miner calls its own co-located inference engine. The gateway strips caller identity and replaces the request ID and nonce with fresh per-dispatch values before anything reaches a miner. See the protocol reference for the exact envelopes.
Roles
| Role | Responsibility | Trust boundary |
|---|---|---|
| Miner (GDPR tier) | Serves admitted EU requests under contract and DPA | Sees plaintext while inferring; admitted through an off-chain process and a signed admission snapshot |
| Miner (permissionless) | Serves validator test items and, after a passing track record, opt-in Peer-to-Peer customer requests | Assumed to be able to log, copy or retain payloads, so it never receives EU-tier requests; customers opt in knowing this |
| Validator | Private schedules, labelled items, identity items, verification, epoch scoring, weight submission | Receives no customer content and no organic metadata; injects probes only through the gateway; must be independent of other validators |
| Gateway / router | Authenticates callers; enforces tier, eligibility and capacity; projects payloads; dispatches; measures latency; records receipts | Customer API keys and chain keys never reach miners; never reads the emission score |
| Reference processor | Computes reference outputs p_ref on byte-identical payloads | Never stores payloads; must be admitted like a GDPR operator for any tier whose organic traffic it shadows |
| Decision Models (external owner) | Customer ingress, tenant policy, legal notices, billing | Owns the decision to send real traffic; REAX cannot override that gate |
| Subnet owner / trust root | Chain hyperparameters, signed trust root, policy, catalog and chain profile | An offline owner key signs issuer keys |
The chain provides hotkey registration, validator permits and stake-weighted consensus over weights. It does not inspect models, prove EU location, bind a company to a DPA, or judge whether answers are correct. Those are the jobs of admission, the gateway and validators.
Tiers
| GDPR tier | Permissionless tier | |
|---|---|---|
| Wire value | gdpr_eu | global_permissionless |
| Who can serve | EU-established legal operators processing data in admitted EU countries | Any chain-registered hotkey that passes conformance |
| How you get in | Off-chain KYB, contract and DPA, then a signed admission snapshot; chain registration is necessary but never sufficient | Chain registration plus a signed miner record and conformance |
| Traffic today | Never sent to the network | Validator test decisions, and opt-in Peer-to-Peer requests from Decision Models (single-question s1-fast only) |
The Peer-to-Peer tier of Decision Models is live. A customer opts an API key in; then single-question s1-fast requests may be answered by REAX miners that have a passing track record at our validator. Every network answer is checked against our own reference GPU. EU-tier requests never go to the network.
Current status
| Area | State | Detail |
|---|---|---|
| Mainnet | Live · netuid 92 | Netuid 92, live since 14 October 2026; Registration is open; the chain sets the burn cost, so check it before you register. |
| Validator | Live | Our validator is UID 3 on an RTX PRO 6000 (96 GB); rounds about every 2 minutes |
| Pool s1-fast | Release 1.2 | Plumb-4B on one NVIDIA GPU with 24 GB or more. Several nodes per person are allowed (one hotkey and UID each) |
| Pool s1-pro | Release 1.2 | mammoth-1, MIT, on an NVIDIA GPU with 48 GB or more; only RTX PRO 6000 Blackwell Server Edition measured. Run the self-check on other cards. |
| Pools s1-lite, s1-cpu | Release 1.2 | 8-bit GGUF on llama.cpp: s1-lite on an NVIDIA GPU with 8 GB or more incl. Jetson Orin (not yet calibrated on real hardware), s1-cpu on an x86-64 AVX2 or ARMv8.2 CPU. Capacity-based scoring applies inside every pool |
| Competitions | Both tracks paused | Generalist paused since 30 October 2026, 16:30 UTC; REAX Engine paused after its Friday slot. No next crowning scheduled. Since the first weight set at or after Monday 2 November 2026, 14:30 UTC, inference gets 30% and follows measured customer demand; it got 40% from the first weight set after the Engine slot on 30 October until then, and 50% before that. REAX Engine remains at 20% and generalist at 30%. Unpaid portions are burned. See Competition mining. |
| Peer-to-Peer tier | Live, opt-in | Decision Models customers can opt API keys in; EU-tier requests never go to the network |
| Testnet | None | No REAX testnet subnet is assigned; practise on a local chain with the kit |
| Images | Disabled | Disabled at the network miner pending isolated decode validation |
| Design specification | Provisional | The protocol, scoring and validator pages follow the earlier design, spec revision 0.6; release 1.2 is the current mechanism. Constants are provisional |
| Source code | Public | reax-subnet, reax-miner-kit, reax-engine; images on ghcr.io/reaxlabs |