Ant Group Open-Sources Ling-3.0-flash-Fin and FinFIRST Benchmark for Financial Agents

On September 3–4, Ant Group's Ant Ling team open-sourced Ling-3.0-flash-Fin, the first finance-enhanced model in the Ling family, with weights available on Hugging Face under an MIT license, alongside FinFIRST, an open benchmark for financial search agents. Both are now publicly available: developers can obtain a finance-domain model at low cost and evaluate it on financial search tasks in an open, reproducible way.

Confirmed

• The model has 124B total parameters with only 5.1B active, uses an efficient MoE architecture, supports a 256K context window, and was built by continually training Ling 3.0 flash on high-quality financial data.
• It is based on the Bailing V3 architecture, with a chat template supporting a controllable thinking toggle and function calling.
• The model is designed for real financial workflows and can be connected to tools such as search, Python, databases and spreadsheets.
• FinFIRST was built by Ant Group with professional support from CICC's investment banking team; V1 includes 123 expert-written tasks, 701 atomic evaluation criteria and 12,300 rubric points, targeting financial search agents.

Why it matters

• Both @FellMentKE and @alifcoder see the combination of an open model plus a public evaluation framework as the key value: the open model lowers the barrier to finance-domain LLMs while the benchmark enables open, reproducible measurement of financial search capabilities.
• FinFIRST's expert- and practitioner-built tasks and scoring system closely mirror real financial workflows, providing a fine-grained reference standard for evaluating financial agents.
• The MIT license further eases commercial use and downstream development.

2026-09-03 ~ 2026-09-04 · 9 related posts

• Ling-3.0-flash-Fin Weights Released: 124B Params, 5.1B Active, 256K Context — Bestlife73 · 2026-09-03
• [source] Ant Group Open-Sources Finance-Enhanced Model Ling-3.0-flash-Fin with 124B Parameters — niacolhealth · 2026-09-03
• Ling open-sources 124B finance model Ling-3.0-flash-Fin alongside FinFIRST benchmark — alifcoder · 2026-09-04
• Ant Group releases FinFIRST, an open benchmark for financial search agents with 12,300 rubric points — FellMentKE · 2026-09-04
• FinFIRST details: 123 expert tasks, 701 atomic criteria and 12,300 rubric points for financial search agents — FellMentKE · 2026-09-04
• Ant Ling releases Ling-3.0-flash-Fin weights on Hugging Face — FellMentKE · 2026-09-04
• [source] inclusionAI open-sources finance-tuned Ling-3.0-flash-Fin: 124B, MIT license — AdinaYakup · 2026-09-04

2 near-duplicate retellings: FellMentKE · FellMentKE