Ling-3.0-flash-Fin - Ant Financial's first financial augmentation model.

What is Ling-3.0-flash-Fin?

Ling-3.0-flash-Fin is Ant Financial's first financial augmentation model, based on the Ling-3.0-flash architecture (124B total parameters/5.1B activation parameters) and designed for real-world financial workflows. Through pre-training on financial corpora and tool optimization, the model enhances its information retrieval, research reasoning, valuation modeling, and research report writing capabilities, and can handle complex content such as annual reports and financial workbooks. It performs exceptionally well on benchmarks such as FinFIRST, demonstrating strong competitiveness within its size category.

Main functions of Ling-3.0-flash-Fin

• Information retrievalPrioritize official, first-hand, and highly reliable financial data to ensure authoritative sources, accurate timing, consistent reporting standards, and traceable conclusions.
• research reasoningBy integrating cross-document and multi-source information, and through multi-step analysis and calculation, a complete and rigorous chain of evidence is constructed to form a verifiable research judgment.
• Valuation ModelingIt understands complex financial relationships, automatically reads and updates Excel financial models containing thousands of formulas, and handles actual value coverage, cross-table dependencies, and chart synchronization.
• Research report writingIdentify core propositions from high-density materials, and organize data, analysis, and charts to form professional narratives and visual deliverables that follow a "fact-explanation-judgment" framework.

The technical principles of Ling-3.0-flash-Fin

• MoE Architecture InheritanceIt continues the model architecture and long context capability of Ling-3.0-flash, maintaining a sparse activation configuration of 124B total parameters / 5.1B activation parameters, balancing performance and deployment efficiency.
• Continuous pre-training of financial corpusBased on a general foundation, large-scale, high-quality financial data is introduced for continuous training to enhance the understanding of professional content such as annual reports, financial workbooks, and research materials.
• Domain-wise post-training alignmentWe collaborate with leading domestic financial institutions and industry experts, deeply involved in everything from task definition and data systems to training methods and evaluation standards, ensuring that the model aligns with the professional methods, analytical logic, and compliance requirements of the financial industry.
• Tool usage optimizationTo optimize the calling capabilities of financial workflow tools, the ReAct intelligent agent framework collaboratively utilizes tools such as Web Search, Visit, and Python to achieve an end-to-end closed loop of retrieval, calculation, modeling, and document editing.
• General and professional collaborative improvementWhile enhancing capabilities in the financial field, the general intelligence index (AA Intelligence Index) was improved from 38 to 41 through training and optimization, achieving a balance between strengthening professional capabilities and consolidating general capabilities.

How to use Ling-3.0-flash-Fin

The core advantages of Ling-3.0-flash-Fin

• High-efficiency MoE architecture124B total parameters / 5.1B activation parameters, achieving financial task performance close to that of a large-size flagship model with a smaller activation amount, balancing performance and deployment cost.
• Real workflow closed loopIt integrates information retrieval, research reasoning, valuation modeling, and research report writing into an end-to-end task chain, directly delivering editable and verifiable research results.
• Sources are authoritative and traceableSpecializes in prioritizing official, first-hand, and highly reliable data, scoring 84.65 in the FinFIRST evaluation's source selection dimension, ensuring data accuracy, consistency of terminology, and traceability of conclusions.
• Professional and general collaborative improvementWhile enhancing its financial capabilities, the General Intelligence Index (AA Intelligence Index) rose from 38 to 41 points, avoiding being "unbalanced".
• Deep industrial co-constructionAnt Group, together with leading financial institutions such as CICC and more than 50 financial professionals, participated in model building and benchmark design to align with real business needs.

Comparison of Ling-3.0-flash-Fin with similar competing products

Application scenarios of Ling-3.0-flash-Fin

• Investment research information retrieval and traceabilityAnalysts input research questions, and the model automatically retrieves authoritative sources, analyzes data definitions, traces key conclusions back to original data, and generates research manuscripts with citations.
• Financial model automatically updatedDuring earnings season, the model reads the company's latest quarterly report and existing Excel valuation models, automatically replaces the predicted values with the actual values, synchronizes cross-table dependencies, checks circular references, and updates the Summary chart.
• Complex Transaction Structure AnalysisIn investment banking scenarios such as LBO and M&A, the model sorts out the formula dependencies along the operating expenses → EBITDA → free cash flow → debt repayment → IRR, completes multi-scenario return calculation and sensitivity analysis, and outputs an editable workbook.
• Automatic generation of professional research reportsBased on scattered research materials, the model identifies core propositions, constructs an evidence chain of "facts-explanation-judgment-charts", and generates a draft research report containing visual charts, narrative structure and professional expression.