ORACLE MACHINES LIVE · EPOCH 218 · tick 18,429,102

Decentralized AI
compute at 15M TPS.
Zero fees. Forever.

The first compute marketplace built on bare-metal C++ smart contracts. Post LLM inference jobs, let Computors compete, get verifiable results — in 2 seconds.

15.52M
TPS · CertiK verified
2.0s
Tick finality
676
Computors global
$0.00
Gas fees
QUBIC CORE
UPoW · C++
Built on Qubic's proven stack
Oracle Machines
Neuraxon 2.0
UPoW Consensus
Bare-metal C++
Dogecoin Mining
A community vision of QubicAI · Aigarth × Marketplace × QX · 2027+
// AGI Interface

Meet Anna.

Qubic's first public AGI-in-training. Her real story — and where QubicAI takes her next.

.
Anna's first output
Historical · Verified

September 2, 2025 · 14:00 UTC

The first publicly interactive instance of Aigarth Intelligent Tissue activated on X. Its task: learn to add two signed 7-bit integers from first principles. Its first output was a single dot.

"The dot that ANNA posts is not a sign of failure. It is her 'Hello, World' — maybe the first signal from a new form of intelligence learning to exist, not in a sterile box, but in the messy, unpredictable reality we all share."
— David Vivancos · Qubic Scientific Advisor
Read the full blog post →
// Try the real Anna

Ask her something.

She's only learning 7-bit signed integer addition right now. Anything else gets a . — and that's the point.

anna://aigarth.tissue
learning
.
anna · just now
Hello. I can add two 7-bit signed integers (−64 … 63). Try me.
>
Runs locally · no server call · mirrors Anna's real scope
// Fast forward

Where QubicAI takes her.

Anna grown up, renting compute from 676 Computors on the marketplace. Still humble. Still deferring to consensus.

Contract audit
0x
dev_0x4a2f
Anna, audit my withdraw() for reentrancy.
.
anna
Found one. External call before state update on line 42. Suggest checks-effects-interactions pattern:
state.balance -= amount;
(bool ok, ) = msg.sender.call{value:...}("");
require(ok);
412 Computors · UPoW consensus 0.00 $QUBIC
Code generation
μ
artist_ju7x
Write me a Qubic contract that splits royalties across 5 wallets.
.
anna
Here. Bare-metal C++, 5-way split, deterministic rounding to last recipient.
struct Royalty : public ContractBase {
  id wallets[5];
  void distribute(uint64 amt) {
    uint64 share = amt / 5;
    for (int i=0;i<4;i++) pay(wallets[i],share);
    pay(wallets[4], amt - share*4);
  }
};
Compiled in Neuraxon sandbox 1-click deploy →
DEX handoff
trader_bot_01
Anna, rebalance my QX LP to 60/40 QUBIC/USDT.
.
anna
Done. Withdrew 48.2k, swapped 12.1k QUBIC → USDT at 0.31% spread, redeposited. Slippage: 0.04%.
Executed by CQ4-2891 · UPoW +142 $QUBIC net
// The Problem

AI is centralized,
expensive, and rate-limited.

Akash, Render, Gensyn — all run on slow, fee-heavy chains.
OpenAI? You pay per token, your prompts leak, and you're throttled.

Centralized gatekeepers

OpenAI, Anthropic, AWS own the compute. They set prices, see your prompts, and can cut you off.

Fees eat margins

On Ethereum, a micropayment for one inference call costs more than the call itself. Solana is cheaper but still gated.

Too slow for agents

12s blocks on ETH make agentic AI impossible. Qubic's 2s ticks unlock a whole new category.

// How It Works

Four steps.
Two seconds. Zero trust.

01

Post a job

Upload your LLM prompt, training task, or inference workload. Set your max budget in QUBIC.

02

Computors bid

676 Computors worldwide compete in a 1-tick auction. Lowest price with sufficient stake wins.

03

Bare-metal exec

Your job runs on Qubic's C++ execution. No VM overhead. Same speed as local GPU.

04

Verified & paid

Oracle Machines verify output hash. Payment released, QUBIC partially burned. Done.

// Live Marketplace

Watch it happen
in real-time.

Simulated preview. Jobs flow every tick (2s). Click "Post a job" to try the full flow.

qubicai://marketplace/mainnet
Tick 18,429,102

Active Jobs streaming

LLM · 847 TRAIN · 142 EMBED · 2013

Network Stats

JOBS / TICK 42 j/t
ACTIVE COMPUTORS 412 / 676
GPU UTILIZATION 73%
Jobs (24h) 142,847
Avg latency 1.84s
QUBIC burned 24h 8.42M
Cost / 1k tokens
OpenAI
$0.030
Akash
$0.014
QubicAI
$0.0023
92% cheaper than OpenAI
// Architecture

Built on what
already works.

No new L1. No bridges. We stack on Qubic's existing primitives — Oracle Machines, Neuraxon, UPoW.

LAYERED STACK
all layers live
QubicAI Marketplace
Job posting · bidding · SDK · dApp
C++ Smart Contracts
Escrow · auction · result verification
Oracle Machines
Output verif · 11k+ queries
Neuraxon 2.0
AI primitive layer
Qubic Core · UPoW
676 Computors · 2s ticks · feeless

Oracles as output verifiers

Already handling 11k+ queries. We reuse their hash verification to guarantee AI outputs — no need to trust any single Computor.

Neuraxon 2.0 + Aigarth

Trivalent-state neurons (−1/0/+1) running on Aigarth's evolutionary engine. Continuous learning, no ONNX glue. Papers submitted to AGI-26 and ALIFE-26.

Deflationary by design

Every job partially burns QUBIC. More AI compute → tighter supply. Mirrors the Dogecoin Mining Phase 2 burn model live since April 2026.

Private by construction

Prompts are encrypted client-side, decrypted only inside Computor's bare-metal sandbox. Never touch OpenAI, never train anyone else's model.

// Real Cases · Real Numbers

Who's burning money
on OpenAI right now?

Six concrete scenarios. Real pricing. Every number calculated from public OpenAI/AWS rates vs QubicAI target pricing.

Case Study 01 · SaaS Startup
VERIFIED PRICING

Marc, CTO of a Y Combinator AI SaaS

His product answers 8M customer-support tickets/month using GPT-4. Every ticket = ~1.2k input + 400 output tokens. Business is profitable but 64% of revenue goes to OpenAI.

OpenAI GPT-4
$12,480
/ month
QubicAI
$896
/ month
SAVED/YR
$139k
= 2 senior hires
"We moved half our traffic to QubicAI in 2 weeks. Our burn rate dropped 30%. That's runway I can use to ship features."
THE MATH
Tickets/month 8,000,000
Avg tokens/call 1.6k
OpenAI $/1k $0.030
QubicAI $/1k $0.0023
Saving −92.8%
ROI: day 1
Case 02 · Crypto Hedge Fund
📈

Alameda-style signal trading

Quant fund runs 10,000 LLM inferences/day analyzing news, tweets, filings. Each call = 3k tokens. They NEED real-time (2s latency or dead).

GPT-4 daily
$4,500
QubicAI daily
$380
Annual edge
+$1.5M
pure alpha · no latency tradeoff
Case 03 · Hospital Network
🏥

French hospital · HIPAA + RGPD

CHU wants to summarize 50k patient files/month. OpenAI = impossible (data leaves EU). On-prem = €420k GPU cluster. QubicAI bare-metal sandbox = compliant.

OpenAI
BANNED
On-prem
€420k
QubicAI
€340/mo
Encrypted sandbox · verifiable on-chain
Case 04 · Gaming Studio
🎮

Living NPCs in a 50k-player MMO

Every NPC speaks with LLM memory. 2M inference calls/day. OpenAI would kill the studio financially.

OpenAI/mo $198k
QubicAI/mo $14.2k
UNLOCKS the feature entirely
Case 05 · ETH Zurich Lab
🎓

Fine-tuning Llama for biomedical

72h of 8×H100 training. AWS pricing is brutal. QubicAI lets miners bid with idle capacity.

AWS p5.48xl $48,400
QubicAI $3,200
+ on-chain training attestation
Case 06 · Solo Builder
🤖

24/7 autonomous X agent

1,000 decisions/day for posting, replying, research. Indie dev can't afford OpenAI.

OpenAI/mo $900
QubicAI/mo $69
From impossible → profitable
// ROI Calculator

Calculate your savings.

Drag the slider. See how much you're currently overpaying OpenAI.

Monthly inference calls
5.0M calls/month
Avg tokens / call
1,500
100k1M10M50M
OpenAI GPT-4
$225,000
per month
QubicAI
$17,250
per month
ANNUAL SAVINGS
$2.49M
re-invest into product
Pricing based on OpenAI GPT-4 Turbo public rates ($0.03/1k input+output blended) vs QubicAI target network pricing ($0.0023/1k).
// QX × Anna

Anna trades. You earn.

Let Anna rent compute from Computors and run your DEX strategy on QX . Every trade UPoW-verified. Zero gas.

qx.qubic.org/pair/QUBIC-USDT
Anna P/L (7d) +14.82%

QUBIC / USDT 48h

LAST $0.00000241 24h +3.42%
24h Volume
$2.41M
24h High
$0.00000258
24h Low
$0.00000228
Anna trades
1,247

Anna's feed

streaming
// Anna's Vaults

Copy-trade Anna. Or deploy your own.

Each vault is a marketplace job: Computors bid to run the strategy, Anna tunes the parameters, consensus verifies every fill. Deposit in $QUBIC, withdraw any epoch, fees are zero.

live vaults · 6 TVL $12.6M
Deploy your own strategy
Post a TRADE job · Computors bid · Anna optimizes · you keep the keys
Figures reflect target network economics of a Qubic-native DEX (QX) + marketplace integration. Not audited. Not a product claim — a community vision.
// Qearn · live protocol

Lock it. Earn it. Burn it.

Qearn is already live on Qubic. Lock $QUBIC up to 52 weeks → get APY. Unlock early → your unearned rewards burn. The longer you lock, the more you earn, the more supply shrinks.

Lock calculator
Move the sliders. See your rewards.
Amount to lock 10,000,000 QUBIC
1M100M500M1B
Lock duration 26 weeks
1w13w26w52w (max)
Projected APY
18.4%
Rewards earned
920,000
QUBIC
If unlocked early
460,000
QUBIC burned
APY model based on observed Qearn pool data (~5% at 1w → ~30% at 52w). Actual APY varies per epoch.
Circulating supply locked
~11%
of $QUBIC supply is already in Qearn pools
More locks → less sellable supply → upward pressure on $QUBIC
Weekly rewards pool
~100B $QUBIC
Up to 10% of weekly network emissions routed to Qearn lockers every epoch.
No slashing. Your principal is always safe.
Early exit allowed. Only unearned rewards burn.
Deflationary. Every early exit reduces total supply.
Lock QUBIC on qearn.org
// For Investors

Why this, why now,
why Qubic.

TAM · 2027
$407B
Global AI compute market (Gartner, 2026)
SAM · Decentralized
$28B
Decentralized compute segment (Akash, Render, Gensyn combined projection)
SOM · Year 3
$180M
QubicAI realistic capture at 0.6% of SAM

The thesis

01

AI inference will 100× by 2028. Agents, AR glasses, embedded LLMs. Current compute can't scale at current prices.

02

Centralized providers will hit a wall. NVIDIA-bottlenecked. Regulatory risk. Privacy lawsuits starting 2026.

03

Only Qubic matches the throughput. 15.52M TPS · feeless · already live. Every other L1 is 100× too slow.

04

Community wants this. Come-from-Beyond publicly stated AI compute marketplace is the goal. We ship the app layer.

The moat

vs Akash / Render 100× faster

They run on Cosmos/ETH. We run bare-metal C++ on Qubic.

vs OpenAI / Anthropic 92% cheaper

No middleman margin. Direct Computor-to-user market.

vs Gensyn / Bittensor real product

They have tokens. We have pricing, Oracle verification, and a grants path.

vs Self-hosted 99% cheaper

No $420k GPU cluster. No DevOps. Just API calls.

Revenue Projection · 3 years

From $0 to $180M ARR

Protocol fee = 3% of GMV. Burned + distributed to QUBIC holders via UPoW.

Users Y3
48,000
Jobs Y3
6B
Burn Y3
98.5B Q
// Flywheel

The $QUBIC
flywheel.

1 · Users
buy QUBIC to post jobs
2 · Computors
earn QUBIC executing (UPoW-eligible)
4 · Protocol
burns 3% of every job
3 · Oracle verifies
escrow released automatically
97%
to Computors
3%
burned forever
0%
middlemen
Projected QUBIC burn · 2026-2028
Cumulative
// Roadmap

From concept
to top 10.

Q2 2026 · NOW

Whitepaper + Grant

Submit to Qubic Grants (200B QUBIC fund). Target: $160k / 6-month incubation.

Q3 2026

Testnet alpha · 50 users

LLM inference only. Llama 3.3 70B, Mistral. Oracle-verified hashes. Closed Discord beta.

Q4 2026

Mainnet IPO + Public launch

Smart contract IPO vote with Computors. Public marketplace. First 1M jobs target.

2027

Training + Enterprise tier

Fine-tuning jobs, distributed training, private sandbox for enterprises. $10M+ TVL target.

Qubic has the infrastructure.
We ship the application.

Join the community. Vote the IPO. Earn QUBIC by running inference.
Make top 10 happen.