Italianhype/Blum-Finance-4B

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 28, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

BLUM Finance 4B by Italianhype is an open, 4-billion parameter financial reasoning model based on the Qwen3-4B architecture. It is specifically designed for evidence-bound financial analysis, providing balanced bull/bear cases, explicit risks, and conservative confidence assessments. This model excels at structured financial reasoning and is intended for use in analyzing financial evidence rather than direct trading or price prediction.

Loading preview...

Overview

BLUM Finance 4B is a specialized financial reasoning model developed by Italianhype, built upon the Qwen/Qwen3-4B base architecture. It is designed to provide evidence-bound financial analysis, focusing on structured reasoning rather than market prediction or trading. The model incorporates a LoRA delta, fused into the base architecture, and is released in a portable BF16 Transformers format.

Key Capabilities

  • Evidence-Bound Financial Reasoning: Structures analysis around balanced bull/bear cases, explicit risks, invalidation conditions, and conservative confidence.
  • Governed Continual Memory: Ships with an installable blum-finance package for schema-validated inference, auditable local memory, and explicit opt-in community contributions, ensuring local retrieval of source-verified outcomes.
  • Robustness: The original MLX release achieved 96.26% aggregate task-contract score, 100% structured validity, and 92.86% no-fabrication on the BLUM temporal reasoning test.
  • Financial MMLU Performance: Achieved 74.86% accuracy on a stratified community evaluation of 700 finance/business MMLU questions, with subjects like Marketing scoring 92% and High-school microeconomics 87%.

Intended Use Cases

  • Financial Analysis: Ideal for analyzing financial evidence, identifying risks, and assessing confidence in financial scenarios.
  • Research and Development: Useful for developers and researchers building applications that require structured financial reasoning capabilities.
  • Educational Tools: Can be integrated into tools for learning and understanding financial analysis methodologies.

Limitations and Safety

  • Not a Price Oracle or Broker: The model is not intended for direct price prediction, trade execution, or as a source of market alpha.
  • Validation Required: Users must validate company identity and market facts against authoritative data sources, as documented fabrication failures (e.g., misidentifying NVDA) can occur.
  • Safety Guidelines: Do not use as unattended financial advice, for authorizing transactions, or without supplying timestamped evidence and validating factual statements.

Finetunes
2 models