Skip to main content

Ant Group Launches Finova Benchmark for Financial AI

Ant Group Unveils Finova Benchmark and Agentar-Fin-R1 Model

July 28, 2025 — At the World Artificial Intelligence Conference, Ant Digital, the technology arm of Ant Group, officially launched the Finova Large Model Financial Application Evaluation Benchmark and its new Agentar-Fin-R1 financial reasoning model. Developed in partnership with China Industrial Bank, Ningbo Bank, and other institutions, Finova sets a new standard for evaluating AI in financial applications.

A Leap in Financial AI

The Agentar-Fin-R1 model, built on the Qwen3 framework, has demonstrated superior performance on authoritative benchmarks like FinEval1.0 and FinanceIQ, surpassing both general-purpose and specialized financial models. According to Ant Digital CEO Zhao Wenbiao, this advancement bridges the 'knowledge gap' between generic AI models and industry-specific needs.

Image

Addressing Industry Challenges

With digital transformation accelerating in finance, AI models must handle complex reasoning, high-level financial knowledge, and strict compliance requirements. Existing models often fall short in real-world financial tasks. Agentar-Fin-R1 addresses these gaps through:

  • A trillion-scale financial data corpus
  • Expert-annotated long-term thinking chains (CoT)
  • Innovative weighted training algorithms

Image

Key Innovations

The model comes in multiple versions (8B to 72B parameters) to suit diverse deployment needs. A MOE architecture variant based on the Bai Ling model enhances inference speed. Notably, Agentar-Fin-R1 can continuously update with new financial policies and market data while reducing fine-tuning costs for enterprises.

The Finova Benchmark

The collaborative Finova benchmark evaluates three critical areas:

  1. Intelligent agent capabilities
  2. Complex reasoning performance
  3. Security compliance standards

Fully open-sourced, Finova aims to elevate industry-wide AI application standards. In initial tests, Agentar-Fin-R1 outperformed even larger general-purpose models.

Image

Real-World Impact

Ant Digital has already deployed its technology across major Chinese banks:

  • Serves 100% of state-owned banks
  • Powers an AI mobile banking app that increased user activity by 25%
  • Developed over 100 financial agent solutions

The company plans further expansion into新能源 (new energy) sectors while continuing to refine its financial AI offerings.

Key Points:

  • Ant Digital launches Finova benchmark with leading financial institutions
  • New Agentar-Fin-R1 model excels in financial reasoning tasks
  • Solution reduces deployment costs through efficient training algorithms
  • Open-source benchmark promotes industry-wide AI advancement

Enjoyed this article?

Subscribe to our newsletter for the latest AI news, product reviews, and project recommendations delivered to your inbox weekly.

Weekly digestFree foreverUnsubscribe anytime

Related Articles

Sakana AI's Tiny Plugin Could Revolutionize How AI Handles Massive Documents
News

Sakana AI's Tiny Plugin Could Revolutionize How AI Handles Massive Documents

Tokyo-based Sakana AI has unveiled groundbreaking technologies that could solve large language models' notorious 'memory anxiety.' Their Text-to-LoRA and Doc-to-LoRA systems enable AI to digest lengthy documents in under a second, shrinking memory requirements from gigabytes to mere megabytes. This breakthrough promises to make customizing AI models dramatically cheaper and more accessible.

February 28, 2026
AI InnovationMachine LearningNatural Language Processing
Chinese AI Models Outpace US Competitors in Global Adoption
News

Chinese AI Models Outpace US Competitors in Global Adoption

In a surprising shift, Chinese AI models have overtaken their US counterparts in global usage for the first time. Platforms like MiniMax and Moonshot AI are leading the charge, with Chinese models accounting for over 5 trillion weekly tokens - nearly double American offerings. This milestone reflects China's growing influence in artificial intelligence development.

February 27, 2026
AI CompetitionChinese TechMachine Learning
Ant Digital Plants Flag in Malaysia Amid AI Expansion Wave
News

Ant Digital Plants Flag in Malaysia Amid AI Expansion Wave

Chinese fintech giant Ant Group is making strategic moves abroad as global demand for AI solutions surges. Its digital arm Ant Digital recently launched an operations hub in Malaysia to strengthen regional services. Meanwhile, the company continues innovating domestically with new enterprise AI offerings. This comes amid explosive growth across the AI sector, with major players like Palantir and OpenAI reporting staggering revenue increases.

February 26, 2026
AI expansionAnt GroupDigital finance
Moonshot AI's Kimi K2.5 Achieves Remarkable Profitability Milestone
News

Moonshot AI's Kimi K2.5 Achieves Remarkable Profitability Milestone

Moonshot AI's latest model, Kimi K2.5, has stunned the tech world by generating more revenue in its first 20 days than all of 2025 combined. The breakthrough comes primarily from overseas users and developers embracing its API services, propelling the company's valuation past $10 billion. Founder Yang Zhilin confirms the company is well-funded with no immediate IPO plans.

February 24, 2026
Artificial IntelligenceTech StartupsMachine Learning
News

Chinese AI Models Capture Global Spotlight During Lunar New Year

Chinese artificial intelligence models made waves internationally during the 2026 Spring Festival, capturing over 60% market share on OpenRouter's developer platform. Three domestic models - MiniMax M2.5, Kimi K2.5, and Zhipu GLM-5 - dominated the rankings by offering superior coding and automation capabilities at remarkably low costs. Their success highlights China's growing influence in AI productivity tools.

February 24, 2026
Artificial IntelligenceChinese TechDeveloper Tools
Google's Gemini 3.1 Pro Outshines Competitors With Breakthrough Reasoning Skills
News

Google's Gemini 3.1 Pro Outshines Competitors With Breakthrough Reasoning Skills

Google has unveiled Gemini 3.1 Pro, its most advanced AI model yet, showcasing remarkable improvements in logical reasoning and problem-solving. The new architecture delivers more than double the performance of its predecessor in critical tests, even surpassing GPT-5.2 in some benchmarks. Beyond raw power, Gemini 3.1 Pro introduces innovative multimodal capabilities, handling ultra-long contexts and generating visual representations of complex concepts.

February 24, 2026
AI InnovationGoogle TechMachine Learning