Ayansk11/FinSenti-Qwen3.5-9B

VISIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Apr 11, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

FinSenti-Qwen3.5-9B by Ayansk11 is a 9.0 billion parameter Qwen3.5-based model fine-tuned for financial sentiment analysis. It excels at classifying short financial texts (headlines, earnings snippets) into positive, negative, or neutral, providing a reasoning chain for each classification. This model is optimized for parsing financial text and generating structured outputs, requiring approximately 20 GB of GPU memory for unquantized operation.

Loading preview...

FinSenti-Qwen3.5-9B: Financial Sentiment Analysis with Reasoning

FinSenti-Qwen3.5-9B is the largest model in the FinSenti collection, a series of models developed by Ayansk11 for financial sentiment analysis. This 9.0 billion parameter model is specifically fine-tuned to interpret short financial texts and provide a clear sentiment classification (positive, negative, or neutral) along with a detailed reasoning chain.

Key Capabilities

  • Financial Text Classification: Accurately classifies short financial texts (1-3 sentences) such as news headlines and earnings snippets.
  • Chain-of-Thought Reasoning: Generates a concise reasoning explanation for its sentiment decision, enhancing transparency and interpretability.
  • Structured Output: Adheres to a strict <reasoning>...</reasoning><answer>...</answer> output format, facilitating easy downstream parsing.

Training and Performance

The model was trained using a two-stage recipe: Supervised Fine-Tuning (SFT) on the FinSenti dataset (~15.2K samples) followed by Generative Reinforcement Learning with Proximal Optimization (GRPO). GRPO utilized four reward functions (sentiment correctness, format compliance, reasoning quality, output consistency), achieving a mean reward of approximately 3.50 / 4.0 on the validation set. It was built upon Unsloth's pre-quantized mirror of Qwen/Qwen3.5-9B.

Use Cases and Limitations

This model is ideal for automated financial news analysis and sentiment tracking. However, it has specific limitations:

  • Short Documents Only: Not designed for long documents; training context was capped at 2048 tokens.
  • Single-Asset Reasoning: Classifies sentiment for individual text snippets, not aggregated across multiple sources.
  • No Numerical Reasoning: Interprets sentiment from numerical mentions (e.g., "beats by 12%") but does not perform mathematical calculations or forecasting.
  • English Only: Trained exclusively on English data.
  • Limited Background Knowledge: Relies on its base pretraining for general knowledge and cannot look up external information.
  • Three-Class Output: Provides only positive, negative, or neutral labels; not suitable for multi-class or continuous sentiment scores without further processing.