Ant Group has released Ling-3.0-flash-Fin as an open-source model, a 124-billion-parameter Mixture-of-Experts system built specifically for investment research and financial analysis. The move gives finance teams direct access to a model co-developed with financial institutions, with open weights that allow private deployment and integration with internal tools.
The model activates 5.1 billion parameters per token, a design Ant said cuts running costs while maintaining the breadth needed for complex financial tasks. It was built with input from industry experts who shaped its task design, data systems and evaluation methods - a response to workflows where traceable sources and auditable outputs are not optional.
Built for four financial workflows
Ant structured the system around four core areas of financial work. Information retrieval prioritises official and authoritative sources so data can be checked from origin to output. Research reasoning combines information from multiple sources and formats into verifiable evidence chains. Valuation modelling handles complex Excel links and supports automated updates while keeping files editable. Report generation pulls together facts, calculations and charts into professional research documents.
The open-weight release lets firms connect the model to external tools - search, Python, databases and spreadsheets - rather than relying on standard interfaces. That matters for teams that need to adapt AI to existing internal processes, not the other way around.
Benchmark results and a new evaluation standard
Ant reported competitive results across several financial benchmarks, including FinFIRST, FinSearchComp Verified, FinCRAFT, FinanceAgent v1.1/v2, APEX-Agents, SpreadsheetBench v1/v2 and Ο3-Banking. These tests span search, spreadsheet handling, reasoning and agent-based financial tasks.
Alongside the model, Ant open-sourced FinFIRST, a benchmark developed with the investment banking team at China International Capital Corporation. FinFIRST V1 contains 123 expert-authored tasks, 701 atomic criteria and 12,300 rubric points. It assesses the full research process, not just whether a final answer matches a reference output. Financial work often demands that users show where information came from, how calculations were made and whether each step can be reviewed later. The benchmark reflects that reality.
Part of a broader domain-specific strategy
Ling-3.0-flash-Fin sits within Ant's wider Ling 3.0 portfolio. The base model, Ling-3.0-flash, is a hybrid-reasoning Mixture-of-Experts system. Other variants include Ling-3.0-tiny for local deployment without cloud dependence, Ling-3.0-flash-VL for image and video inputs, and Ling-3.0-flash-SantΓ© for healthcare and life sciences. The pattern is clear: specialised models for sectors where domain-specific requirements dictate how systems are evaluated and deployed.
The release adds to a growing trend among large technology groups to ship domain-focused systems with accompanying benchmarks. In finance, that trend has increasingly centred on tools that support research, analysis and document workflows while preserving traceability and control over data handling. For professionals building or buying AI for Finance, the signal is that performance should be measured across the full chain of work, not just final-answer accuracy.
Why this matters for finance professionals
Open-weight models like Ling-3.0-flash-Fin change the calculus for in-house AI adoption. Instead of sending sensitive financial data to third-party APIs, teams can deploy the model privately and connect it directly to proprietary spreadsheets, databases and research tools. The co-development with financial institutions also means the evaluation criteria - traceability, auditability, step-by-step reasoning - were baked in from the start, not bolted on later. For CFOs and finance leaders mapping out their AI strategy, this kind of domain-specific release is worth watching. An AI Learning Path for CFOs can help leadership teams assess where models like this fit into forecasting, risk analysis and reporting workflows - and where they do not.
Your membership also unlocks:
