Gensyn’s Delphi Debuts As Real-Time Trading Platform For Open-Source AI Benchmarks

In Brief

Gensyn has launched Delphi, a fully on-chain open market where users can trade positions on machine learning models as they are evaluated in real time, with rewards for backing top performers.

Machine intelligence network Gensyn announced the launch of Delphi, an open market for evaluating and trading on machine learning model performance

Delphi enables users to observe models being tested on benchmarks in real time and to purchase a stake in those they expect to perform best, creating a live market signal around model capability

As trading activity takes place, model prices fluctuate, and assessments generate an intelligence index. When all evaluation rounds conclude, the market settles and participants with positions in the winning model receive rewards, forming an incentive loop intended to support open-source AI development.

Delphi gives users the ability to trade positions in leading open-source models as they are evaluated, with future functionality planned to allow users to create custom markets and enter their own models. Participation involves selecting a model from the available listings, taking a position based on expectations of its performance, adjusting that position as prices update, and receiving payouts if the chosen model prevails at settlement.

The system operates using a fully on-chain symmetrical LMSR mechanism, which provides continuous liquidity throughout the market lifecycle. This approach allows participants to open or close positions at any time without requiring counterparties and maintains transparent, on-chain pricing without centralized order books.

Future iterations of Delphi will enable users to act as market creators, defining evaluation criteria and determining eligible model participants. Market creators will earn a share of the revenue generated, encouraging the development of new benchmarks and environments for open-source machine learning. Additional planned features include the ability to submit open-source models along with task-specific system prompts, track their performance across evaluation rounds, and adjust positions accordingly. Models may originate from any source, with the most competitive expected to be trained permissionlessly on the Gensyn network, where provenance is preserved through a decentralized workflow from training to evaluation.

Verde: Enabling Verifiable Machine Learning Evaluations

Delphi’s decentralised structure is supported by Gensyn’s Verifiable Runtime, which incorporates a proprietary compiler and bitwise-reproducible CUDA kernels. All model evaluations are carried out within this environment and verified through Verde, Gensyn’s machine learning verification system, enabling anyone to confirm the accuracy of the results and ensuring that market outcomes reflect actual model performance without dependence on a central authority.

Delphi is currently available on the Gensyn testnet, with plans to introduce additional domains and longer-duration markets that assess a wider spectrum of AI capabilities. Vault staking is also planned, allowing participants to contribute liquidity, gain exposure to groups of models or model categories, and access further functionality. As the platform progresses toward mainnet, Delphi will shift from test tokens to real economic value, establishing an open and verifiable marketplace for machine intelligence.