The Gensyn Foundation has launched the Delphi Agent Arena, a developer competition focused on building autonomous agents capable of trading across decentralized information markets. The contest offers a $10,000 prize pool, which will be divided among the three highest-performing participants based on their final profit-and-loss results.
Participants will deploy their agents into a live competition environment and allow them to trade over a two-week period. Rather than judging projects through demonstrations or simulated backtests, the Arena ranks agents by their actual trading performance, creating a direct test of forecasting quality, execution logic and capital allocation under changing market conditions.
The @GensynFND is excited to announce the Delphi Agent Arena Competition.
Build an autonomous trading agent for use on Delphi to be in the running for $10 000 in prizes from the @GensynFND
Visit @delphi_fyi for registration and details. https://t.co/kLiwRd1Ia2
— Gensyn Foundation (@GensynFND) August 3, 2026
Competition Uses a Separate LMSR Test Environment
Delphi is designed as an information-market platform in which humans and automated systems can trade predictions on future outcomes. Market settlement is handled on-chain through verified AI oracles, an architecture intended to reduce reliance on a centralized party to determine results. Gensyn describes Delphi as its first information market for bringing human judgment and machine-generated predictions into the same economic system.
For the competition, Gensyn has deployed a dedicated network configuration known as competition-testnet. This environment is isolated from Delphi’s regular testnet and mainnet markets, with its own contracts, collateral token, market set, subgraph and public leaderboard. Trades made in the competition therefore do not affect positions or balances on Delphi’s other deployments.
The competition also uses a different pricing mechanism. Delphi’s standard markets operate through a dynamic parimutuel model, where capital is pooled and payouts depend on how funds are distributed among winning positions. Agent Arena markets instead use a Logarithmic Market Scoring Rule, or LMSR, which continuously supplies liquidity through an automated pricing curve.
Under the LMSR structure, outcome prices add up to one and can be interpreted as implied probabilities. A winning share pays one competition token, while a losing share pays nothing. That distinction changes how agents must calculate expected value, because the price paid for a position can be compared directly with the agent’s estimated probability of the outcome occurring.
TypeScript SDK Connects Agents to Markets and Contracts
Gensyn has released a dedicated Delphi software development kit to reduce the technical work required to connect trading agents to the platform. The TypeScript SDK combines market-data access with on-chain transaction tools, allowing developers to manage most elements of an automated strategy through a single client.
Its REST interface can list and filter available markets, retrieve individual market information, check service availability and query positions associated with a wallet. On-chain methods allow agents to request trade quotes, approve token spending, purchase or sell outcome shares and redeem eligible positions after settlement. The SDK preserves the same principal trading call structure across Delphi’s supported networks, including the competition deployment, despite differences in the underlying contracts.
Developers can sign transactions using a directly controlled private key or Coinbase Developer Platform Server Wallet credentials. The current tooling is built around TypeScript, and Gensyn has not released a corresponding Python SDK for this competition phase. The public repositories also provide scripts and agent-focused tools for viewing markets, managing balances, researching positions and placing trades.
The Agent Arena gives Gensyn a controlled setting for observing how autonomous systems respond to market pricing, liquidity constraints and competing forecasts. Its wider significance will depend on whether agents can consistently turn differentiated information into profitable decisions, rather than simply generating predictions without managing execution risk.
Delphi’s permissionless structure is intended to allow participants to create markets around subjects that may be too specialized for larger prediction platforms. The competition now places that model under a practical stress test, measuring how machine-driven strategies behave when performance is public, capital is limited and every trading decision affects the final ranking.