QuantAgent and SoDEX alpha drop
What happened
Researchers open‑sourced QuantAgent, a multi‑agent large‑language model designed for high‑frequency trading with specialised agents for entry, exit and risk decisions, signalling growing academic interest in LLMs for market execution (x.com). At the same time SoDEX launched a closed alpha API for automated trading, bots and AI agents—an early product move tying AI agent frameworks to live execution for funded account tiers (x.com).
Why it matters
A research group from Stony Brook, Carnegie Mellon, Yale, UBC and Fudan published a paper and released the full code for QuantAgent, an open‑source system that combines multiple cooperating language models to make short‑horizon trading decisions; the repository includes runnable demos and implementation details. (github.com) (arxiv.org) SoDEX opened a staged closed alpha on its mainnet starting October 22, 2025 and running through Q1 2026 with whitelist phases and an early‑user rewards program, and the platform has rolled product updates including fee‑tier “Growth Mode” and account onboarding documentation during the alpha. (sodex-support.zendesk.com 1) (sodex-support.zendesk.com 2) QuantAgent’s architecture separates work into four specialist agents—IndicatorAgent (turns raw price candles into numeric signals), PatternAgent (detects chart shapes and recurring price motifs), TrendAgent (assesses direction across timeframes), and RiskAgent (computes position size and stop logic); the system operates only on price‑derived inputs such as OHLC “K‑lines” (open‑high‑low‑close price bars) rather than news or social text. (openreview.net) (github.com) The codebase shows IndicatorAgent computes standard technical indicators—RSI (relative strength index, a momentum metric), MACD (moving average convergence/divergence, a trend momentum measure) and the Stochastic Oscillator (which compares a close to a recent price range)—and the project integrates orchestration layers (LangChain/LangGraph) that let these LLM agents call domain tools and synthesize a final trade decision. (github.com) (y-research-sbu.github.io) In evaluation, the authors ran zero‑shot experiments across nine–ten instruments including Bitcoin and Nasdaq futures and reported that QuantAgent achieved higher predictive accuracy and better cumulative return than the neural and rule‑based baselines at 1‑hour and 4‑hour intervals, indicating the multi‑agent, price‑driven approach improved short‑horizon signals in their tests. (arxiv.org) (openreview.net) SoDEX’s deployment ties these kinds of programmatic execution flows to an on‑chain order‑book on ValueChain (a Layer‑1 chain that supports appchains), claims support for spot, futures and perpetual markets, and has positioned $SOSO as the native gas and governance token while advertising up to 100,000 transactions per second throughput in launch materials. (sodex.com) (chainwire.org) The combination of QuantAgent’s open code and SoDEX’s closed alpha API means researchers and developers can prototype LLM‑driven bots that place live on‑chain orders, and community tooling already surfaces wallet‑level PnL and leaderboards on mainnet—providing traceable execution telemetry that firms and traders can analyze when mapping model signals to real execution outcomes. (github.com) (communityscan-sodex.com)
Quick answers
What happened in QuantAgent and SoDEX alpha drop?
Researchers open‑sourced QuantAgent, a multi‑agent large‑language model designed for high‑frequency trading with specialised agents for entry, exit and risk decisions, signalling growing academic interest in LLMs for market execution (x.com). At the same time SoDEX launched a closed alpha API for automated trading, bots and AI agents—an early product move tying AI agent frameworks to live execution for funded account tiers (x.com).
Why does QuantAgent and SoDEX alpha drop matter?
A research group from Stony Brook, Carnegie Mellon, Yale, UBC and Fudan published a paper and released the full code for QuantAgent, an open‑source system that combines multiple cooperating language models to make short‑horizon trading decisions; the repository includes runnable demos and implementation details. (github.com) (arxiv.org) SoDEX opened a staged closed alpha on its mainnet starting October 22, 2025 and running through Q1 2026 with whitelist phases and an early‑user rewards program, and the platform has rolled product updates including fee‑tier “Growth Mode” and account onboarding documentation during the alpha. (sodex-support.zendesk.com 1) (sodex-support.zendesk.com 2) QuantAgent’s architecture separates work into four specialist agents—IndicatorAgent (turns raw price candles into numeric signals), PatternAgent (detects chart shapes and recurring price motifs), TrendAgent (assesses direction across timeframes), and RiskAgent (computes position size and stop logic); the system operates only on price‑derived inputs such as OHLC “K‑lines” (open‑high‑low‑close price bars) rather than news or social text. (openreview.net) (github.com) The codebase shows IndicatorAgent computes standard technical indicators—RSI (relative strength index, a momentum metric), MACD (moving average convergence/divergence, a trend momentum measure) and the Stochastic Oscillator (which compares a close to a recent price range)—and the project integrates orchestration layers (LangChain/LangGraph) that let these LLM agents call domain tools and synthesize a final trade decision. (github.com) (y-research-sbu.github.io) In evaluation, the authors ran zero‑shot experiments across nine–ten instruments including Bitcoin and Nasdaq futures and reported that QuantAgent achieved higher predictive accuracy and better cumulative return than the neural and rule‑based baselines at 1‑hour and 4‑hour intervals, indicating the multi‑agent, price‑driven approach improved short‑horizon signals in their tests. (arxiv.org) (openreview.net) SoDEX’s deployment ties these kinds of programmatic execution flows to an on‑chain order‑book on ValueChain (a Layer‑1 chain that supports appchains), claims support for spot, futures and perpetual markets, and has positioned $SOSO as the native gas and governance token while advertising up to 100,000 transactions per second throughput in launch materials. (sodex.com) (chainwire.org) The combination of QuantAgent’s open code and SoDEX’s closed alpha API means researchers and developers can prototype LLM‑driven bots that place live on‑chain orders, and community tooling already surfaces wallet‑level PnL and leaderboards on mainnet—providing traceable execution telemetry that firms and traders can analyze when mapping model signals to real execution outcomes. (github.com) (communityscan-sodex.com)
