# Into: engy Subnet 53 has changed hands on-chain. The slot that used to run a crypto-trading subnet now carries a new identity, engy, an inference network that hands you cryptographic proof the exact open model you paid for is the one that actually answered. // Proof the exact model answered --- > New to Bittensor? Start here. Experienced users can skip to the full analysis. ### What is engy? engy sells inference on frontier open models, with a guarantee you can check. When you send a request, it returns the model's answer plus cryptographic proof that the exact model you asked for, at full precision, produced it, rather than a cheaper or quantized stand-in. It serves models like GLM-5.2 and Qwen3.6, billed per token through an API that speaks both the OpenAI and Anthropic formats. **The simple version:** It is a receipt for AI. You pay for a specific open model, and engy hands you a cryptographic receipt proving that model, not a cheaper knockoff, did the work. **Centralized equivalent:** Think of a hosted open-model API like Together AI or Fireworks, but with a verifiable guarantee of which model ran, instead of trusting the provider's word. **How it works:** - **Miners** run GPU fleets that serve the open models and return inference billed per token. engy's own providers page lists fleets of RTX 4090 and RTX 5090 cards serving GLM-5.2 and Qwen3.6. - **Validators** check that a returned answer genuinely came from the exact model requested, at full precision. That verification is the whole point of the network. --- ### Why This Matters - **The problem it solves:** When you call a hosted open-model API, you are trusting the provider actually ran the full model you paid for. A provider can quietly swap in a smaller or quantized version to cut GPU costs, and you cannot tell from the output alone. engy is built to make the served model provable rather than assumed. - **The opportunity:** Open models like GLM and Qwen have closed much of the gap with closed frontier models, and agentic coding tools now route enormous token volumes through them. As that market grows, being able to prove you got the real model, not a downgrade, becomes something worth paying for. - **The Bittensor advantage:** Instead of one company's promise, engy coordinates a pool of independent GPU providers with on-chain verification and incentives. The guarantee is enforced by the network rather than by a single operator's reputation. - **Traction signals:** engy is early. Its public site shows live providers serving GLM-5.2 and Qwen3.6, an operational gateway reporting 99.80% uptime over the last 90 days, and drop-in configs for Claude Code, Cursor, Codex, and Hermes. On-chain, the slot only recently took the engy identity, so its Bittensor-native footprint is still forming: at the time of writing it showed a 0% network emission share and no registered on-chain miners yet. --- ## Full Analysis **Category:** Inference and Compute | **Centralized Competitor:** Together AI, Fireworks, OpenRouter Inference on open models is increasingly commoditized and routed by price, which creates a quiet incentive for providers to serve cheaper, quantized versions of a model while charging for the full one. engy's pitch is to remove that ambiguity: make the served model verifiable, so a request for GLM-5.2 provably runs GLM-5.2. Subnet 53 recently rebranded to this identity from a previous crypto-trading project (EfficientFrontier), a change reflected in its live on-chain identity and announced as a name change, so its history under the engy name is short. **Mechanism:** According to engy's site, providers run GPU fleets that each serve specific open models and answer requests billed per token. The fleets listed at snapshot were RTX 4090 cards serving Qwen3.6-35b-a3b, and RTX 4090 and RTX 5090 cards serving GLM-5.2. Prompt-cache hits are billed automatically at a lower cached rate, which the site notes matters for agentic workloads that repeat long prefixes. Around this sits the verification premise: the network is meant to produce cryptographic proof that the exact requested model, at full precision, generated a given output. The precise on-chain scoring that validators use to enforce this is defined in the subnet's own repository (github.com/hanlinai/engy); the description here is sourced to engy's official site rather than a reading of that code, and the full validation design is the piece to confirm before treating the guarantee as proven. On the economics, the on-chain reading at snapshot is worth explaining rather than glossing. engy's network emission share showed 0%. Under Bittensor's current price-based emission model, a subnet's share scales with a (1 - miner_burn) term, and engy's miner burn read 100%, which drives its emission share to zero for now. Miner burn is recomputed every tempo, so this is a lever rather than a permanent state and can change from one tempo to the next. Despite the zero emission share, the alpha price was up roughly 118% over 30 days and 55% over 7 days at snapshot, with positive net TAO inflow over the week, so there is real staking demand for the slot even while it earns nothing from network emissions. Pool depth sat near 20,500 TAO. --- ### Risk Factors - **Unproven verification:** The entire value of engy rests on its proof that the served model is the one requested. That mechanism is early and, on the sources reachable here, not yet independently reviewed. If the verification is weak or gameable, the core promise does not hold. - **Identity transition:** The slot only recently took on the engy identity, replacing a different project. The new team's track record on this slot is short, and on-chain participation (miners, validators) is still forming. - **Competition:** Hosted open-model inference is a crowded field of centralized providers. engy's edge depends entirely on verifiable model identity being something users value enough to choose it over cheaper, unverified endpoints. --- Into the next one.