Ant Group open-sources the financial model Ling-3.0-flash-Fin with 124 billion parameters.

ME News reports that on September 4 (UTC+8), Dongcha Beating AI announced that Ant Group’s Ling-3.0-flash-Fin, a financial large language model, has been officially open-sourced. The model features 124 billion total parameters and 5.1 billion activated parameters, with support for 256K context length. The official BF16 weights, approximately 255 GB, are released under the MIT license and can be self-deployed using vLLM and SGLang. Alongside the model, Ant Group has also open-sourced the financial search benchmark FinFIRST. Developed by Ant Group with professional support from the investment banking team of China International Capital Corporation (CICC), over 50 financial experts participated in question creation and validation. Only 9.78% of candidate questions passed screening, resulting in a final set of 123 questions. Of these, 61.8% require calculations, and 32.5% require cross-referencing multiple sources. Each question is broken down into multiple scoring criteria to evaluate whether the AI correctly identifies data, uses appropriate sources, verifies timeframes and statistical methodologies, and produces accurate calculations. The full dataset contains 701 granular scoring points. The questions, reference answers, and scoring rubrics are now publicly available under the Apache 2.0 license. (Source: BlockBeats)