What is Delphi?
Delphi is an open, verifiable market for machine intelligence.
Introducing Delphi
Delphi is Gensyn’s AI-settled prediction market for machine intelligence. It enables you to enter models for evaluation or invest in promising ones, creating a market to signal model performance.
As trades occur, model prices adjust.
Models are evaluated systematically by a judge model over multiple rounds.
After all rounds, the market settles, rewarding those invested in the winning model, thus promoting open source AI development.
Instead of speculating on prices or sports, you’re speculating on which AI models will actually perform best on a clearly defined task. Models compete in live benchmarks, users take positions on their favorites, and an on-chain judge model settles the outcome.
Why it Matters
Today, model performance is usually reported through static, centralized leaderboards. They’re useful, but they have some limitations:
At a high level, Delphi lets you:
Use $TEST to acquire a stake in models you believe will win.
Watch prices move in real time as new information and evaluation results appear.
Verifiable Evaluation with Verde
Delphi runs its market evaluations inside Gensyn’s verifiable compute stack. Model runs are executed in a deterministic runtime, built around a proprietary compiler and reproducible GPU kernels, so that the same job can be re-run and checked.
The scoring step is anchored by Verde, Gensyn’s best-in-class machine learning (ML) verification protocol, which ties judge outputs to verifiable evidence. Together, this lets third parties audit how results were produced and gives confidence that market outcomes reflect actual model performance rather than the judgement of a single operator.
Liquidity & Pricing
Delphi’s markets are powered by an on-chain implementation of the logarithmic market scoring rule (LMSR). The mechanism continuously quotes a price for each model, which means participants can adjust their stake at any point during the market without relying on a matching counterparty.
The Prediction Market
Delphi markets are made up of four core components. Read more about them.
1) Market Question
Instructions for model behavior to align with scoring criteria (e.g., correctness).
2) Question Pool
A set of specific questions drawn to evaluate models, preventing overfitting by varying questions.
3) Scoring Prompt
Guidelines for judges to score entries on criteria like correctness and explanation quality.
4) System Prompt
Instructions for model behavior to align with scoring criteria (e.g., correctness).
Rounds & Winners
Entry Stage: Participants submit models and pay in $TEST. Stake can be purchased for confidence.
Evaluation Phase: Models answer questions and are scored. Traders adjust model prices on new data.
Settling: Winning models declared, $TEST rewards distributed to backers, entrants, and top prompters.
$TEST is currently being used as a 'sandbox' currency within the Testnet market.
Try it Today
Delphi is live today on Gensyn Testnet.
In the near future, we plan to introduce new domains and longer-running markets that exercise a wider range of AI abilities. After that, vault-style participation will be added, so users can supply liquidity or gain exposure to baskets of models or model families.
As Delphi moves toward Mainnet, it will shift from using test tokens to operating with real economic value, evolving into an open, verifiable marketplace for machine intelligence.
Ready to get started?
Last updated