precisionalgorithms/qwen3.5-9b_precision_agentic_trading
The precisionalgorithms/qwen3.5-9b_precision_agentic_trading model is a 9 billion parameter LoRA fine-tune of Qwen/Qwen3.5-9B, optimized for precision agentic trading. It excels at tool use for financial platforms like Robinhood, Base, and Coinbase, featuring a 32768-token context length. This model is designed to read tool schemas, build arguments with correct units, gather data before acting, and execute decisions without requiring re-confirmation. It achieves significantly improved autonomous task completion (75.0%) and structured output parse rate (98.3%) compared to its base model, making it suitable for local, self-directed trading agent deployments.
Loading preview...
Model Overview
This model, qwen3.5-9b_precision_agentic_trading, is a 9 billion parameter LoRA fine-tune of the Qwen/Qwen3.5-9B base model, specifically engineered for agentic trading applications. It is designed to interact with financial platforms such as Robinhood, Base, and Coinbase by interpreting tool schemas, constructing precise arguments, and executing actions autonomously. The model operates with a 32768-token context length.
Key Capabilities & Performance
- Enhanced Tool Use: Significantly improves autonomous task completion to 75.0% (from 45.0%) and multi-step sequencing to 58.2% (from 0.0%).
- Structured Output: Achieves a 98.3% structured output parse rate, ensuring reliable communication with tools.
- Precision & Integrity: Demonstrates 100% unit precision and 98.8% required argument matching.
- Autonomous Execution: Reads tool schemas, builds arguments with correct units, gathers data, and executes without requiring re-confirmation.
Noteworthy Limitations
- Refusal Rate: The model refuses approximately 18% of valid orders, a known defect where it declines rather than acting incorrectly. This is a trade-off for its improved precision and safety.
- Tool Selection: Tool selection performance is worse than the base model, primarily due to its tendency to refuse action rather than calling the wrong tool.
Ideal Use Case
This model is best suited for users who want to run their own trading agent locally, with custom prompts and strategies, without sending their positions or intent to a hosted model. It acts as an execution assistant, verifying call formation and data trustworthiness, then performing instructed actions without offering market opinions or investment advice.