Run open models on Bittensor subnet 64, where miners work inside sealed hardware and can't read your words, or on a GPU someone lends, or on your own. Mine with your GPU, and send sealed messages between wallets. Pay with your own Chutes key, or with credits on Robinhood Chain when payments are on.
Yes. Start mining from the Lend page and your GPU answers people on Inferno's network, right in your browser. You keep 70% of what each answer costs when people pay with credits on Robinhood Chain. That split is planned.
Qwen2.5 7B on a lent RTX 3090. 164 tokens at 58 tokens/s. $0.0016, settled on Robinhood Chain.
Is there a much bigger model that keeps my question private?
Question: Is there a much bigger model that keeps my question private? Answer: Yes, on Bittensor. A miner on subnet 64 answers with a big open model like DeepSeek-V3.2, inside sealed hardware, so the miner can't read your words. Chutes' gateway passes them along in memory and doesn't keep them. Receipt: Served on Bittensor subnet 64 (Chutes), confidential compute. DeepSeek-V3.2, 54 tokens at 42 tokens/s. $0.0001, billed to your Chutes account.
Question: What does a GPU do when I ask you something? Answer: It does the maths that picks each next word. A model is billions of numbers, and a GPU multiplies thousands of them at once. That's why a gaming card can write faster than you can read. Receipt: Llama 3.1 8B on a lent RTX 4090. 212 tokens at 86 tokens/s. $0.0021, settled on Robinhood Chain.
Question: Can you answer without sending my question anywhere? Answer: Yes. Switch to private mode and I run on your own GPU, inside this browser tab. Your question and my answer never leave your device, and it costs nothing. Receipt: Llama 3.2 1B on your GPU (Apple M3), in private mode. 96 tokens at 24 tokens/s. Free, and nothing left this tab.
Question: I have a gaming PC. Could it earn? Answer: Yes. Start mining from the Lend page and your GPU answers people on Inferno's network, right in your browser. You keep 70% of what each answer costs when people pay with credits on Robinhood Chain. That split is planned. Receipt: Qwen2.5 7B on a lent RTX 3090. 164 tokens at 58 tokens/s. $0.0016, settled on Robinhood Chain.
Listening for Robinhood Chain…
Every spike on those lines is a real Robinhood Chain block, about ten a second.
Live on Bittensor subnet 64
Asking subnet 64 which models are live…
What powers the answers?Powered by Bittensor.
Sealed
Subnet 64 runs each model inside a sealed enclave. The miner owns the hardware but can't look in.
Bittensor is a decentralized network where miners compete to serve AI. The network pays them in TAO, its own coin.
Inferno uses subnet 64, run by Chutes. Every model it lists there runs inside confidential-compute hardware, a sealed enclave that stops the miner reading your prompts.
Asking subnet 64 which models are live…
Two ways to pay
Your own Chutes keyLive
Paste a key from chutes.ai and answers bill your Chutes account, which takes TAO too. The key stays in your browser and goes only to Chutes.
Inferno creditsWhen payments are on
Send USDG from your wallet to Inferno on Robinhood Chain, or TAO once the operator switches it on. Each answer costs Chutes' price plus 20%, our planned margin. The credits page shows what this server takes today.
How does it work?One question, one GPU, one payment.
You ask
Type in the chat. In private mode, your question never leaves your device.
A GPU answers
A Bittensor miner's, on subnet 64, inside sealed hardware, so the miner can't read your words. A lender's, picked by wallet address, in the network beta. Or your own, through WebGPU.
The chain pays
Payment per token, in USDG, settles on Robinhood Chain. The lender keeps 70% (planned).
Why is the chat different?It shows the work behind every answer.
Heat trace
Words arrive hot and cool as they settle. The faster the GPU, the hotter they burn.
Receipts
Every answer ends with what it cost, which model wrote it and where it ran.
Private mode
The model runs on your own GPU. Nothing leaves the tab, and it's free.
Model choice
SmolLM2 360M, Llama 3.2 1B and Qwen2.5 1.5B run in your browser, or on a lender's GPU in the network beta. For much bigger models, like DeepSeek, Qwen3.5 397B, Kimi and GLM, switch to Bittensor.
A token is a chunk of text, often part of a word. Models read and write in tokens, and that's what you pay for: about four tokens for every three words.
Model
Llama 3.2 1B
Ran on
Your GPU, Apple M3, in private mode
Speed
24 tokens/s
Tokens
61
Cost
Free. Nothing left this tab.
Can my GPU earn?Mine with your GPU.
Mining here means your GPU answers people's AI questions and gets paid for the work. There are two ways in, depending on the GPU.
GPU utilisation
IdleServing requests
Illustrative: a mining GPU heating up as questions arrive.
Outside the browser
Mine on Bittensor
For data-center GPUs, on your own servers, as a miner on subnet 64.
What you do
Run a Chutes miner. It serves open models to everyone who uses Chutes, Inferno's Bittensor chat included.
What you earn
Subnet 64's alpha token, which swaps for TAO, paid by the Bittensor network rather than Inferno. Validators score every miner, so earnings follow how well yours serves.
What you need
A server of data-center GPUs (Chutes tests 8× H200, B200 or RTX Pro 6000) running as Intel TDX confidential VMs, a Bittensor wallet registered on subnet 64, and TAO for the registration fee.
Is this live?Chat, mining and credits are. Sealed messages are next.
Robinhood Chain seals a block about ten times a second and settles it to Ethereum. Every number here is read from it live.
Inferno uses it for money only. You buy credits by sending USDG, and lenders are paid in USDG for the tokens their GPUs write. Your prompts and replies never go on-chain; only the payments do.
Listening for Robinhood Chain…
Latest block
–
Average block time
–
Transactions per second
–
Base fee
–
Ethereum-final after
–
The last 24 blocks, newest on the right. Taller, hotter bars carried more transactions.
What does it cost?Pay per token. Private mode is free.
Network models charge for the tokens they read and write, and nothing else. No subscription, no minimum.
Pay with Inferno credits: send USDG on Robinhood Chain and your balance shows up once the chain finalizes it, in about 16 minutes. It doesn't expire.
The lender whose GPU answers gets 70% of what you pay. During the beta, lender payouts go out by hand in USDG.
2,000words a day
20020,000
Bittensor modelChutes' price for the model, plus 20% when you pay with credits
Varies by model
Private modeRuns on your GPU, in your browser
Free
Network modelA lender's GPU, $0.01 per 1,000 tokens
$0.80 a month
About 80,000 tokens a month. Models count text in tokens, roughly 1.33 per word.
Is it private?It depends on the mode. Here's exactly who sees what.
What each mode reveals about your words, your wallet, timing and payments
Bittensor chat
Private chat
Network chat
Sealed messages
Your words
Bittensor chat: Chutes' gateway, in memory, and Inferno's server in passing when you pay with credits. Miners can't read them.
Private chat: Only you
Network chat: You and the lender whose GPU serves the request
Sealed messages: You and the recipient
Your wallet
Bittensor chat: No one with your own key; Inferno's server when you pay with credits.
Private chat: No one. No wallet needed.
Network chat: Inferno's server, to bill your credits. Your top-ups are public on Robinhood Chain. Lenders never see it.
Sealed messages: The recipient, and the relay that routes the envelope
Who and when
Bittensor chat: Chutes sees request time and size, and so does Inferno when you pay with credits.
Private chat: No one
Network chat: Inferno sees request time and size to bill you
Sealed messages: The relay sees sender, recipient and time, never content
Payments
Bittensor chat: Your Chutes account, or Inferno credits on Robinhood Chain.
Private chat: Nothing to pay
Network chat: Public on Robinhood Chain, like any transfer
Sealed messages: Free during beta
Bittensor chat, private chat and network chat work today. Sealed messages still run on a demo relay that keeps everything in your browser.
You can message people too.
Notes between wallets are sealed on your device, with your key and Ada's, before they leave your screen. The relay only ever carries the sealed box. It can see which keys are talking and when, but not a word you wrote, or even how long it was.
Your safety number with Ada
Ada sees the same 60 digits on her screen. Read a few groups to each other once. If they match, nobody swapped a key between you.
You and Ada are demo keys made in this tab just now. Nothing leaves your browser.
Sealed as you type, from your key to Ada's.
The relay sees
Two keys, a time and the sealed bytes. No names, no words.
From
… your key
To
… Ada's key
Sent
…
Nonce
… random, used once
The sealed note: 272 bytes that only Ada's key can open.
You wrote 94 bytes. Sealed, it's 272, one pattern per byte. Every note up to 252 bytes seals to the same 272, so the relay can't tell “ok” from a paragraph.
Ada opens
Nothing opened yet.
Anything else?Questions people ask
What is Bittensor, and how does Inferno use it?
Bittensor is a decentralized network where miners compete to serve AI, and the network pays them in TAO. Inferno's Bittensor mode asks subnet 64, run by Chutes, where miners serve big open models inside confidential-compute hardware, so they can't read your prompts. Chutes' gateway handles your words in memory and, by its privacy policy, doesn't store them.
Can I pay with TAO?
Once the operator switches it on. TAO is on Robinhood Chain through Chainlink's bridge, and Inferno can accept it for credits; the credits page lists what this server takes today. If you use your own Chutes key, Chutes takes TAO directly.
Is Inferno part of Robinhood?
No. Inferno is independent and built on Robinhood Chain, a public network anyone can build on. It isn't affiliated with or endorsed by Robinhood Markets.
Do I need a wallet?
Not to chat in private mode, or on Bittensor with your own Chutes key. You need one to pay with Inferno credits, to mine with your GPU and get paid, and to send sealed messages.
Which models can I run now?
SmolLM2 360M, Llama 3.2 1B and Qwen2.5 1.5B, in private mode on your own GPU, or on a lender's GPU in the network beta. The first load downloads the model once, from about 400 MB to 2 GB, and your browser keeps it. For much bigger models, like DeepSeek, Qwen3.5 397B, Kimi and GLM, switch to Bittensor mode.
How does mining pay?
On Inferno, your GPU earns 70% of what each answer it serves costs, when people pay with credits on Robinhood Chain. The split is planned, and during the beta payouts are sent by hand, in USDG. On Bittensor, the network pays miners in subnet 64's alpha token, which swaps for TAO; Inferno isn't involved.
What stops a lender from faking work?
Today, less than we'd like. Every lender signs its node key with its wallet, so you know which address served you, and you're never billed for more text than you actually received. Spot checks with known prompts, to confirm each GPU runs the model it lists, are planned.
Is my chat stored?
In private mode your chat stays in your browser, and you can delete it. On the network, Inferno doesn't store prompts, but the lender whose GPU serves a request can read it.
Why Robinhood Chain?
A new block lands about every 0.1 seconds and fees are low, so paying per token is practical. USDG is native to the chain, and it settles to Ethereum.