What is FLock.io?
FLock.io is a decentralized AI training platform that pairs federated learning, where models learn from data that never leaves its owner, with blockchain-based rewards and checks. Participants stake FLOCK to train or validate models in competitions and collaborations, and they earn tokens based on results. Its mainnet and token launched on Base at the end of 2024.
Utility Lens
What FLock.io actually does
Coordinates and rewards people who train, validate and improve AI models while the underlying data stays with its owners.
- Real-world usage
- earlyAI Arena and FL Alliance are live with thousands of models and many participants, but rewards come mainly from daily token emissions, and we found little public data on revenue from paying outside customers.
- Token necessity
- integralStaking FLOCK is required to train, validate or delegate, and rewards are paid in FLOCK, though the trained models can be used without it.
- Decentralization
- limitedTraining and validation are spread across many nodes, but the FLock team still sets task rules and emissions, and its DAO design is described as not yet final.
- Track record
- developingThe project was founded in 2022 and its token and mainnet launched on December 31, 2024.
What it’s used for
In AI Arena, training nodes compete to build the best model for a task while validators score their work, and both stake FLOCK and share daily emissions. FL Alliance lets groups fine-tune models together on private data, and the API Platform serves finished models to apps. At its December 2024 launch FLock reported about 20,000 models created and over 600 training nodes.
Role of the FLOCK token:
Evidence of real use
- FLock launched its mainnet on Base on December 31, 2024, together with the FLOCK token generation event, with the token debuting on Bybit. Source (opens in a new tab)
- Users must stake FLOCK to take part in model creation, training or validation, and dishonest behavior can lead to slashing. Source (opens in a new tab)
- In a 2025 paper, FLock researchers say they fine-tuned a 70-billion-parameter language model across untrusted parties in a decentralized setup, a first by their account. Source (opens in a new tab)
The AI angle
- AI role
- model marketplace
- What actually happens
- FLock pays people in FLOCK to train, validate and fine-tune AI models, then makes the resulting models available to apps through its API Platform. The token is staked to take part and slashed for bad behavior, so it is tied to the training process itself. We found little public data on how much paying model users spend compared with the rewards paid out.
How FLock.io works
Federated learning trains a shared model by sending the model to the data rather than collecting the data in one place. Each participant improves the model on local data and returns only the updates. Normally a central server merges those updates, which creates a single point of trust and an easy target for poisoned updates. FLock replaces that server with smart contracts, staking and peer voting, so participants who submit bad work can be caught and slashed.
AI Arena is a public competition. A task creator posts a goal and dataset, training nodes submit models, and validators score them against hidden data. Rewards depend on each participant's stake and the quality of their work, and token holders can delegate FLOCK to nodes in return for a share of rewards. FL Alliance is the collaborative side, where groups fine-tune the strongest models on their own private data.
Finished models can be served through FLock's API Platform. FLock says it works with the United Nations Development Programme, Alibaba Cloud's Qwen models and the Ethereum Foundation. Emissions are handled by an on-chain task manager that pays trainers, validators and delegators in AI Arena each day. The FLock team says its DAO design, legal structure and on-chain governance tools are still being worked out, so the details may change.
Key moments
- 2024FLock raises a $6 million seed round in March and opens the AI Arena beta in May.
- 2024FLock launches its mainnet on Base on December 31 together with the FLOCK token generation event.
- 2025FL Alliance launches in January for collaborative training, and FLock partners with Alibaba Cloud's Qwen models.
Supply
FLOCK's total supply is capped at 1 billion, and new tokens are minted daily through a management contract rather than all at once. Community incentives are minted over 60 months with a 1% monthly decay. Team and investor tokens, about one third of supply, vest over three years with a one-year cliff.
Risks to understand
- Daily emissions add new supply for several years, and rewards may not be matched by paying demand for trained models.
- Federated learning reduces data exposure but does not remove every privacy or poisoning risk.
- Kraken restricts FLOCK for US clients, so US access depends on fewer venues.
- The DAO and governance design are still unsettled, which leaves key decisions with the core team for now.
FLock.io FAQ
What is federated learning?
It is a way to train an AI model across many devices or organizations without moving their raw data. Only model updates are shared.
What do I need FLOCK for?
You stake it to train or validate models, or delegate it to someone who does, and rewards are paid in FLOCK.
Which blockchain does FLock use?
FLock runs on Base, an Ethereum layer-2 network, where its mainnet and token launched in December 2024.
Sources
- FLock.io Announces Mainnet Launch and Token Generation Event on Base (opens in a new tab) — Decrypt (Chainwire press release)
- FLock.io in 2024: Year in Review (opens in a new tab) — FLock.io
- Token Allocations (opens in a new tab) — FLock.io documentation
- Token Utility: Demand (opens in a new tab) — FLock.io documentation
- Scaling Decentralized Learning with FLock (opens in a new tab) — arXiv
- FLock.io (opens in a new tab) — FLock.io
- Where is Kraken licensed or regulated? (opens in a new tab) — Kraken Support
- Flock price (opens in a new tab) — Coinbase
Last reviewed October 20, 2026. Written by Dave McNaught. Utility Lens ratings are editorial judgments, not investment advice. Market data from CoinGecko. Spot an error? Request a correction.

