The plain-English explainer

Bittensor, explained the way you’d explain it to a friend.

No whitepapers. No math. Just the idea, the money, and the 30 biggest things being built on it right now — in the clearest, simplest language the Bittensor community has produced, so it actually sticks.

Start with the big idea Skip to the 19 highlighted subnets
21M
TAO max supply, like Bitcoin
128
subnets competing for emissions
~$251K
paid to productive subnets daily
19
subnets highlighted below
live
chain data · Aug 28, 2026
01 · The big idea

What is Bittensor, in one breath?

Imagine you want to form a company — but instead of hiring employees, you publish a list of tasks. Freelancers from all over the world compete to finish each task the best. Only the winners get paid. And here’s the catch: you don’t pay them. A Bitcoin-like blockchain pays them. That’s Bittensor.

“Imagine you’re going to form a company, but instead of forming a company, you decide to make a list of tasks. And there are freelancers, some with PhDs, from all over the world who compete to finish each task the best. Only the best freelancers get paid — and you don’t pay them. A Bitcoin-like blockchain pays them. And that’s Bittensor.”
TAO
The coin. Exactly 21 million will ever exist — the same hard cap as Bitcoin. New TAO is created with every block (~every 12 seconds) and paid out to whoever did the most useful work.
A subnet
One contest. Each of the 128 subnets defines a job — answer AI questions cheaply, find DNA mutations, rent GPUs, detect deepfakes — and pays whoever does that job best.
Miners
The workers. Anyone on Earth can plug in a computer and compete — you don’t need permission, a degree, or a Silicon Valley address.
Validators
The judges. They grade the miners’ work and decide who gets paid. Token holders back them with stake, so cheating costs the crowd money.
Emissions
The freshly printed TAO paid out per block. It flows to the subnets the market values most — ~$251,000 per day across the network as of Aug 28, 2026.
“Bitcoin doesn’t have companies. There’s no employees. There’s just the goal — find the hash of the previous block — and whoever does it best wins the Bitcoin every block. Bittensor makes mining programmable. What Bitcoin did for stranded energy, Bittensor does for stranded talent.”

The 30-second version

Bitcoin proved something weird: if you pay people worldwide to compete at one task, you get the largest supercomputer on Earth and $2.5 trillion in value — with no company at all. Bittensor asked: what if that trick worked for AI, science, and every other kind of useful digital work? So it took Bitcoin’s exact economics — 21M coins, halvings, mining contests — and made the task programmable. Each of the 128 subnets is a different task. Together they pay out roughly a quarter-million dollars a day to the people doing the most useful work on the internet.

02 · The machine

How it actually works

Three kinds of players, one block every 12 seconds. Miners do useful work, validators grade it, and the chain pays the winners. The product gets cheaper and better every day — because losers don’t get paid at all.

Diagram: miners and validators connected to a subnet hub which rewards useful work, with users and apps receiving the commodity
Every subnet is a marketplace: miners do the work (green), validators grade it (purple), users get the product (amber), and the subnet rewards the whole loop with new TAO.
“Each subnet has its own sort of Hunger Games contest of miners competing to earn the TAO. The validators are the judges of the Hunger Games — they decide which miners did the best. Only the best and only the fastest get paid. It’s sort of Highlander: there can be only one.
STEP 01

A job is posted

A subnet owner designs a contest — say, “answer AI questions at the lowest price” (Chutes, SN64) or “find mutations in DNA” (Minos, SN107).

STEP 02

The world competes

Anyone, anywhere, plugs in and does the work. No application, no permission — just results. This is “mining.”

STEP 03

Judges grade the work

Validators re-run, benchmark, and verify miner output — bad miners earn nothing and get pushed out.

STEP 04

The chain pays winners

Every block mints new TAO and routes it to the best work. Miners and validators split ~82% (41% each); the subnet owner gets ~18% to run the contest.

Why it beats hiring

“The subnet has 256 temp employees from all over the Earth competing to get my coins. The best brains are fighting against the best brains, and only the top brain gets anything. It’s survival of the fittest, globally. So I’m getting extreme quality, probably a lot cheaper.” — an in-depth Bittensor explainer conversation, Apr 2026

03 · The money layer

dTAO: the chain has its own built-in Subnet Token DEX

In February 2025 Bittensor upgraded to “dynamic TAO.” Every subnet got its own token (called its alpha), and its own built-in Subnet Token DEX — a Uniswap-style pool inside the chain itself — now prices every subnet token. More demand for a subnet → higher price → bigger share of daily emissions.

Diagram: TAO pool on the left, dTAO hub in the middle, subnet rewards on the right, feedback loop beneath
You stake TAO into a subnet’s pool and get its alpha token. The market’s collective opinion of each subnet sets how much TAO that subnet earns per block.
“What’s really interesting is the Bittensor TAO Uniswap is baked directly into the Bittensor chain. The chain itself is the exchange.
“Holding TAO is like owning the S&P 500 of all the subnets. If five or six of them just run and go crazy, it’ll still drag TAO up in value, because TAO sits in the liquidity pool opposite every subnet token. If you don’t know which one to bet on, you just hold TAO.”
FROM JM Crypto, Mar 2026 · figure of speech — subnet tokens are cryptoassets, not stocks or equity
Alpha token
Each subnet’s own coin — same 21M cap as TAO, emitted into its pool. Miners and validators are paid in alpha; stakers hold it. It’s priced directly in TAO.
The pool
The only way to get a subnet’s alpha is to stake TAO into that subnet’s pool — there is no separate crypto exchange account, no insider presale. That’s why subnets had “fair launches.”
Why it matters
Subnet tokens are “rehypothecated TAO” — not economically separate. A breakout subnet pulls TAO up with it; TAO holders fractionally participate in every subnet’s success.”
“Subnet tokens can only be obtained by staking $TAO. There is a direct economic relationship between subnet tokens and $TAO — subnet tokens ARE rehypothecated $TAO. They are different forms of the same thing. This is VERY different from the ‘gas token’ chains like Ethereum or Solana.”
“Anyone anywhere on Earth can be a VC in this world. Unlike venture capital — where your money is bricked for 7 to 10 years — here you wake up one morning, decide you don’t believe in a subnet anymore, swap back to TAO, and you’re out. All brains on Earth coordinate to decide which subnets look best.”
04 · Putting money to work

Staking: the four-step version

You don’t have to be technical to participate. Buying TAO and staking it — into a validator or directly into subnets you believe in — is how most people play. Stakers earn a share of emissions; good picks pay double-digit yields, bad ones dilute you.

Four steps: buy TAO, stake to a validator or subnet, earn emissions, compound or unstake
“It requires TAO to make subnets, to mine, and to stake. Something like 90% of all TAO is staked — it’s not liquid, it’s not moving. And if you’re staked in a subnet, you get a portion of all the emissions that go to it — in some cases yields most people wouldn’t believe.

Nothing here is financial advice

Every good explainer says it on air constantly, and it’s true: subnet tokens are young, volatile, experimental assets. This site explains the machine; it does not tell you what to buy. The happy path — figure out the weird wallet and the weird tools before everyone else does — is exactly how early Bitcoin users described Mt. Gox. Do your own research.

05 · The main event

19 subnets worth knowing, in plain English

The 18 subnets highlighted in the widely-shared subnet primer report — plus #19, Lium, which earned its bonus slot. What each one does in plain English, the traction numbers that matter, and why it’s worth knowing. Live market data from the chain as of Aug 28, 2026.

Minos logo
SUBNET 107

Minos

$25.3Mmarket cap

Decentralized genomic variant calling. HelixForge manufactures realistic practice DNA with answers built in — solving the 'answer key' problem for training mutation-detection AI. OpenAI examined 8 genomics tools and singled out Minos's HelixForge as the most ambitious.

OpenAI genomics report feature
Aug 1, 2026
Rounds
72-min, winner-take-all
Storage
Hippius SN75 + R2 + S3
Whitepaper
20 pages
Worth knowing: OpenAI examined 8 genomics AI tools and singled out Minos's HelixForge as THE most ambitious. Bittensor miners are analyzing DNA and detecting disease mutations — and OpenAI noticed.
139.6 𝜏/day ($32,508)
chain emissions
product
taostats
Engy logo
SUBNET 53

Engy

$39.5Mmarket cap

Verified inference API with cryptographic TOPLOC proof on every response — no TEEs needed. Earth's lowest-cost access to Kimi K3 at exactly half the official price. 4 models live, 28+ miner nodes, 98.87% uptime.

Kimi K3
$1.50/$7.50 per 1M — half official
DeepSeek V4 Flash
$0.045/$0.09 per 1M
Uptime
98.87%
Drop-in
Claude Code, Cursor, Codex, Hermes
Worth knowing: Kimi K3 at HALF the official price — a 2.8-trillion-parameter frontier model running on gaming GPUs, with cryptography proving you got the real model.
29.6 𝜏/day ($6,893)
chain emissions
product
taostats
Actual logo
SUBNET 95

Actual

$36.4Mmarket cap

Private AI inference on your own hardware. One binary install, end-to-end encrypted relay — the company itself cannot read your prompts. 9 models in registry. Native Hermes Agent integration.

Models
9 (DeepSeek V4 Flash, Gemma 4, Qwen3.6, Nemotron)
Relay
end-to-end encrypted
Integration
Native Hermes Agent (Aug 2026)
APIs
OpenAI + Anthropic compatible
Worth knowing: Hermes Agent by Nous Research runs natively on your own Actual-powered hardware — private frontier models with no VPN, no tailscale, no PyTorch babysitting.
early stage
chain emissions
product
taostats
404-GEN logo
SUBNET 17

404-GEN

$12.9Mmarket cap

Text-to-3D generation that democratizes 3D content creation. Web app for prompting 3D models (~1 minute each), a catalog of generated Gaussian splats, and plugins for Blender and Unity.

Generation
~1 minute via web app
Blender plugin
v0.8.0 on Gumroad
Unity plugin
v0.3.0 on Unity Asset Store
Format
Gaussian splats catalog
Worth knowing: Ships downloadable plugins for Blender AND Unity — the two most-used 3D/game engines in the world. A Bittensor subnet wired straight into the workflows of millions of game developers.
9.7 𝜏/day ($2,260)
chain emissions
product
taostats
ReadyAI logo
SUBNET 33

ReadyAI

$6.0Mmarket cap

Agentic data marketplace — llms.txt indexes plus processed podcast datasets, making the web machine-readable. 10K+ domains, 300K+ HuggingFace downloads, 2M+ enrichment points.

Domains
10,000+ indexed
HF downloads
300K+
Enterprise customers
Ipsos, Gelt, Archway
Cost
660x cheaper than MTurk
Worth knowing: 10,000+ domains indexed and searchable in real time, 300K+ HuggingFace downloads, 3 named enterprise customers — at 660x cheaper than Mechanical Turk.
5.0 𝜏/day ($1,173)
chain emissions
product
taostats
Score logo
SUBNET 44

Score

$52.1Mmarket cap

Vision AI research company making every camera intelligent. Its Manako product turns any camera network into real-time 'systems of action' — no new hardware required. Co-selling alliance with PwC France & Maghreb.

PwC co-selling
8-month vetting, enterprise distribution
Paris Blockchain Week
won vs ~1,000 competitors
Projection
$40M ARR (per CEO, Apr 2026)
Mentat Lend
first credit market on Bittensor
Worth knowing: A Big 4 consulting giant (PwC, $57B revenue, 86% of the Fortune 500) is now actively SELLING a Bittensor subnet's tech to enterprise clients. Projecting $40M ARR in 2026.
39.4 𝜏/day ($9,163)
chain emissions
product
taostats
Bitsec logo
SUBNET 60

Bitsec

$6.8Mmarket cap

AI code-vulnerability detection. Its ScaBench benchmark covers 31 real audit projects and 555 bugs from Code4rena/Cantina/Sherlock. Runs free security triage on all ~128 subnets. Delivered a paid audit (MegaTao, Feb 2026).

ScaBench
31 projects, 555 bugs
Coverage
free triage, all ~128 subnets
Paid audit
MegaTao, Feb 2026
Model
audit/bug-bounty marketplace
Worth knowing: Free triage agents already catch Tenex-style exploit classes — and the subnet turned that reputation into its first paid audit. Security as a competition.
0.5 𝜏/day ($120)
chain emissions
product
taostats
Chutes logo
SUBNET 64

Chutes

$106.7Mmarket cap

Leading decentralized serverless AI compute. Trillions of tokens per month, 13 live SOTA models all on Intel TDX confidential compute, new models live within minutes of release. Accepts USD and TAO.

Throughput
trillions of tokens/month
Models
13 SOTA, all Intel TDX
Docs
119 pages
Partners
OpenRouter, Kilo, FetchAI, Phala
Worth knowing: New SOTA models live within minutes of release, at ~85% below AWS pricing — and its Parallax framework proved decentralized training works with only ~0.6% quality loss vs a datacenter.
74.8 𝜏/day ($17,412)
chain emissions
product
taostats
Nova logo
SUBNET 68

Nova

$28.5Mmarket cap

Crypto-native drug discovery platform. Merit-based ML competitions screen a 3.2-billion+ molecule library against 30 protein targets, scored by the PSICHIC oracle. Hits flow to a robotic synthesis lab for real-world validation in days.

Screened
102,932 / 3.2B molecules (live counter)
Targets
30 protein targets
Whitepapers
3 (V1–V3, 2025)
Partners
OnePot.AI (robotic synthesis), Yalotein
Worth knowing: 102,932 molecules screened from a 3.2-billion-compound library, with hits synthesized by OnePot.AI's robotic lab for real wet-lab validation — in days, not months.
22.8 𝜏/day ($5,302)
chain emissions
product
taostats
LeadPoet logo
SUBNET 71

LeadPoet

$6.4Mmarket cap

B2B lead generation finding high-intent buyers from 12+ real-time signal sources. Secured an enterprise sales pilot with Dropbox — validation from a category-defining Fortune 500 company.

Meetings booked
142
Conversion
38 (27% rate)
ARR
~$1M
Pilot
Dropbox (Jul 30, 2026)
Worth knowing: Dropbox is now piloting LeadPoet's AI sales intelligence. 27% conversion rate, 142 meetings booked, ~$1M ARR — real revenue from real enterprise customers.
early stage
chain emissions
product
taostats
Hippius logo
SUBNET 75

Hippius

$22.9Mmarket cap

Decentralized cloud storage: files are split into 30 Reed-Solomon shards across independent miners (any 10 reconstruct your file, tolerating 20 simultaneous failures). Per-object NaCl encryption, S3-compatible API — no wallet required.

Erasure coding
any 10 of 30 shards rebuild
Price
$0.003/GB/month — ~30x cheaper than S3
Egress
free
Apps
desktop app; HippChat demo
Worth knowing: 30x cheaper than AWS S3 with free egress, per-object encryption, files split into 30 shards — and third-party apps (HippChat, Chutes, Minos) already build on it.
19.7 𝜏/day ($4,579)
chain emissions
product
taostats
OpenRoboto logo
SUBNET 80

OpenRoboto

$6.8Mmarket cap

Open robotics VLA-model competition on π0.5/LIBERO. Evaluation reproduces official 96.8–97.2% benchmark scores. Anti-overfitting via 10,030 perturbed task variants — object-position swaps collapse model success 99% → 25%, proving it rewards generalization, not memorization.

Benchmark
reproduces official 96.8–97.2%
Anti-overfit
10,030 perturbed task variants
Generalization test
99% → 25% on position swaps
Stage
early
Worth knowing: Object-position swaps collapse model success from 99% to 25% — proving the subnet rewards generalization, not memorization. Bittensor isn't just software — it's training AI for physical robots.
1.6 𝜏/day ($369)
chain emissions
product
taostats
Yanez logo
SUBNET 54

Yanez

$6.0Mmarket cap

Zero-knowledge proof-of-humanhood and proof-of-uniqueness for the agentic economy. Beta (Q2 2026). Integrations claimed with Google PST, Visa TAP, Stripe MMP, and W3C DIDs — the KYC-free trust layer for AI-agent-authorized payments.

Proof
ZK humanhood + uniqueness
Integrations (claimed)
Google PST, Visa TAP, Stripe MMP
Standards
W3C DIDs
Stage
beta, pre-revenue
Worth knowing: In a world of AI agents, how do you prove you're human? Yanez does it cryptographically — on Bittensor — with Google, Visa, and Stripe integrations claimed.
0.9 𝜏/day ($221)
chain emissions
product
taostats
Bitcast logo
SUBNET 93

Bitcast

$20.0Mmarket cap

Decentralized content marketing expanding beyond Web3 into AI, SaaS, fintech, dev tools, cybersecurity. New products: Outfold (enterprise YouTube marketing with a living intelligence layer) and Stitch3 (real-time X ecosystem intelligence). Now accepting fiat AND crypto.

Campaigns
114 delivered
Creators
880+ onboarded
Audience
10.3M (Web3 alone)
Revenue
$320K → 58% recycled into buybacks
Worth knowing: 114 campaigns, 880 creators, 10.3M audience in Web3 alone. Now expanding to enterprise tech, taking fiat payments — with 58% of revenue recycled into token buybacks.
20.4 𝜏/day ($4,753)
chain emissions
product
taostats
gm logo
SUBNET 28

gm

$27.5Mmarket cap

Private frontier-model gateway (Claude/GPT/Gemini) routed through an Intel TDX secure enclave on Phala. 30+ models at or below maker prices, drop-in for Cursor/Cline/Claude Code. Prompts decrypted only inside the enclave — invisible to the model maker, the host, AND gm itself.

Models
38 live (all TEE-verified)
Privacy
decrypted only inside enclave
Drop-in
Cursor, Cline, Claude Code
Pricing
at/below maker prices
Worth knowing: 38 frontier models — Claude, GPT, Gemini, Qwen 3.5-397B — through ONE API, verified by Intel's silicon. Even the gateway operator can't read your prompts.
20.0 𝜏/day ($4,646)
chain emissions
product
taostats
Targon logo
SUBNET 4

Targon

$79.9Mmarket cap

Bittensor's flagship decentralized AI inference subnet (by Manifold Labs), serving popular open-source LLMs through an OpenAI-compatible REST API. Miners run inference, validators verify responses. Evolved into a Confidential Compute Cloud for GPU/CPU rentals.

GitHub
72★, 61 forks — most-forked subnet repo
API
api.targon.com (OpenAI-compatible)
Backed by
OSS Capital (lead) + DCG
Privacy
TVM confidential compute
Worth knowing: 72 GitHub stars and 61 forks — one of the most community-adopted Bittensor codebases in existence. Backed by DCG and OSS Capital. The oldest living subnet (SN4).
55.7 𝜏/day ($12,968)
chain emissions
product
taostats
Iota logo
SUBNET 9

Iota

$40.4Mmarket cap

Distributed LLM pretraining across permissionless machines, by Macrocosmos. Published arXiv paper. Currently training a 1.5B model, scaling toward 100B. 'Train at Home' app lets anyone with a GPU join the swarm.

Project Orion
LIVE — 256+ distributed GPUs
In training
62B/16B; Orion-100B in progress
Paper
arXiv:2507.17766
Parent
Macrocosmos runs 3 subnets
Worth knowing: Training a 100-BILLION-parameter model across 256 decentralized GPUs that nobody owns — and it works. Published arXiv paper. Train-at-home app for anyone with a GPU.
early stage
chain emissions
product
taostats
Beam logo
SUBNET 105

Beam

$4.5Mmarket cap

Decentralized bandwidth coordination network using Proof-of-Bandwidth. Completed the first decentralized subnet-to-subnet data transfer: 107GB from Cloudflare R2 to Hippius in ~6 minutes. Beam Tunnels connect private enterprise infrastructure securely.

Miners
1,247
Aggregate
847 Gb/s
Volume
2.4 PB
Integrations
S3, R2, GCS, Snowflake, Salesforce, MCP
Worth knowing: The first decentralized subnet-to-subnet transfer in Bittensor history — 107GB from Cloudflare R2 to Hippius in ~6 minutes. Boring, real, billable internet infrastructure.
13.3 𝜏/day ($3,095)
chain emissions
product
taostats
Lium — bonus #19 logo
SUBNET 51

Lium — bonus #19

$110.4Mmarket cap

The #2 subnet on Bittensor by market cap ('Airbnb for GPUs'): 1,286 GPUs live, 874 rented (68% utilization). First agent-first compute provider. Confidential Compute with TEEs live (Jul 2026). Listed on Kraken and MEXC. The most aggressive deflationary tokenomics in the ecosystem.

Daily auto-burn
~$20K/day (1,412 owner tokens per 24h, on-chain verified)
Single buyback-and-burn
$700K, Apr 2026 — 100% revenue-funded
Sept 2025
$3M+ burned
Rank
#1 of ALL subnets for buybacks/burns within 3 days of launch
Worth knowing: Lium burns ~$20,000 of its own token EVERY DAY, on-chain, forever — and ranks #1 of all 128 subnets for buybacks-and-burns. Revenue up, supply down. 874 GPUs actively rented = real paying demand; revenue funds the burns; burns support the price.
85.2 𝜏/day ($19,833)
chain emissions
product
taostats
07 · Pocket dictionary

Every word you’ll hear, defined once

TAO (τ)

The base coin of Bittensor. Hard-capped at 21 million, like Bitcoin. Everything on the network is priced in it.

Subnet (SN#)

One competition on the network — a job (serve AI, find molecules, store files) plus the miners and validators doing and grading it. 128 exist today.

Alpha (α)

A subnet’s own token, capped at 21M, priced in TAO inside the chain’s built-in exchange. You get it by staking TAO into the subnet’s pool.

Miner

Anyone doing the subnet’s actual work — running GPUs, writing models, providing bandwidth. Paid in proportion to how good their work is.

Validator

The graders. They verify miner output and set the payment weights. Big token holders back them with stake, which is how you “stake to a validator.”

Emissions

The new TAO/alpha printed every block (~12s) and paid to winning work. The network’s payroll: ~$251K/day as of Aug 28, 2026.

dTAO

“Dynamic TAO” — the 2025 upgrade that gave every subnet its own token and let the market, not a committee, decide each subnet’s share of emissions.

Subnet Token DEX

The automated market maker (Uniswap-style) built directly into the Bittensor chain. It’s where TAO swaps for each subnet’s alpha token — cryptoassets priced by supply and demand, not shares of a company.

Root network

The base layer of TAO staking — the low-risk, index-like way to earn from the whole network at once instead of picking subnets.

TEE

“Trusted execution environment” — a hardware-sealed vault inside a computer chip. Lets strangers run your job without ever being able to see your data. Used by Targon, gm, Chutes.

Yuma Consensus

The algorithm that turns validators’ grades into payments — designed so a few powerful validators can’t rig the outcome. “Fuzzy, not deterministic.”

Inference

Using an AI model that already exists — asking ChatGPT-style questions and getting answers. “Training” builds the model; “inference” runs it.

DePIN

“Decentralized physical infrastructure network” — crypto-incentivized real-world hardware (GPUs, storage, bandwidth). Most top subnets are DePIN for AI.