Introducing Delphi
Delphi lets you watch machine learning models compete live on benchmarks and buy a stake in those you think are best, creating the first live market signal of model performance.
Today we are introducing Delphi, the open market for machine intelligence.
Delphi lets you watch machine learning models compete live on benchmarks and buy a stake in those you think are best, creating the first live market signal of model performance. As trades are placed, the price of each model changes. Models are systematically evaluated against assessment criteria, producing an index of model intelligence. When all the rounds are complete, the market settles and those with a stake in the winner earn rewards - creating a flywheel that drives development in open source AI.
Click here to participate in the first market - Gensyn Middleweight General Reasoning Benchmark. Delphi allows users to trade positions in top open source models as they undergo evals. Soon you will be able to create your own markets and compete with your own models as well, all secured by Gensyn’s decentralised verifiable machine learning infrastructure.
Buy stake in a model
Users can participate by buying stake in any model:
- Browse the list of entered models.
- Buy stake in the models you believe will win as new information appears.
- Enter or exit at any time as prices adjust.
- Receive payouts if the model you backed wins at final settlement.
Liquidity and pricing
Delphi uses a fully on-chain symmetrical LMSR to provide continuous liquidity throughout each market. This is one of the only live implementations of such a system and allows users to smoothly enter & exit a position at any time without waiting for a counterparty. It allows the market to update cleanly from the first trade to the final settlement, using a fully transparent and on-chain pricing mechanism, no opaque centralised order books.
Create a market (coming soon)
In future versions of Delphi, users will be able to take on the role of market creator, defining the evaluation criteria and the eligible model participants for their own live market. The more interesting the market, the more other people will participate - providing a live ranking of interesting benchmarks and environments for open source models to explore.
Revenue accruing to the market will be shared with the market creator, incentivising the production of new benchmarks, environments, and problems for machine learning to solve.
Enter a model (coming soon)
Also in future versions, users will be able to enter an open source model and a system prompt crafted for the market
- Select the model you want to enter.
- Write a system prompt crafted for the relevant task(s).
- Monitor results as each evaluation round is published. Decide whether to maintain, exit, or add to your position at any time.
- Earn rewards if your model wins the evaluation at settlement.
Models could come from anywhere, but the strongest will be trained permissionlessly on the Gensyn network through the Signal, Scale, Eval flywheel. Models trained in this way will maintain attribution for every resource provided, enabling true decentralised machine learning from inception to evaluation, at unlimited scale.
Verifiable evaluation with Verde (coming soon)
Underpinning the decentralisation of roles in the Delphi system will be Gensyn’s Verifiable Runtime - which includes a proprietary compiler and bitwise-reproducible CUDA kernels. The model assessments will be executed within this runtime and secured using Verde, Gensyn’s best-in-class ML verification system. This allows anyone to verify that the evaluation is correct and ensures markets accurately reflect model performance without relying on a trusted intermediary.
Delphi is live today on Gensyn testnet. Over the coming weeks it will expand with additional domains and longer-running markets that cover a broad range of AI capabilities. Vault staking will follow, allowing users to provide liquidity, buy into baskets of models or model families, and more. As we approach mainnet, Delphi will transition from test tokens to real economic value, creating an open and verifiable market for machine intelligence.
Visit the docs: https://docs.gensyn.ai/intelligence-market/what-is-delphi
Delphi: https://delphi.gensyn.ai/
"Delphi provides open access to machine intelligence markets. Instead of investing in private companies, like OpenAI, market participants can now back open source models directly as they are evaluated technically. Trade real technical progress rather than marketing hype and AGI promises, create new global evals tracking model capabilities, and build new models collaboratively with on-chain attribution."
- Ben Fielding | Co-Founder and CEO
Gensyn networks together the core resources required for machine intelligence to flourish alongside human intelligence, providing a verifiable and standardised way to run ML tasks across any device in the world. The vision is a future where the three product verticals of signal, scale, and evaluation form the full machine learning lifecycle and the backbone of an open AI economy. An economy powered by real usage, collaborative training, and trusted verification; enabling.