Mint-Agent models debut with strong performance on financial benchmarks

Researchers have introduced Mint-Agent, a new family of foundation models specifically designed for financial applications. These models are built on three core components: a specialized data engine for financial tasks, a harness for stable interaction with environments and auditable evidence trails, and a training recipe combining supervised fine-tuning, optimal policy distillation, and reinforcement learning from human feedback. The flagship models, Mint-Cu (9B) and Mint-Ag (27B), demonstrate strong performance on financial benchmarks, outperforming existing models like GPT-5.6-Sol and Claude-Opus-4.8 in reliability and executability. AI

IMPACT Establishes a new benchmark for trustworthy financial AI by jointly engineering domain expertise, long-horizon execution, and auditable evidence.

RANK_REASON The item is a research paper introducing new agentic foundation models for finance. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Agents-A1-35B
• arXiv
• Claude (Opus 4.8)
• FinanceAgentBench v1.1
• FinanceAgentBench v2
• FinSearchComp T2
• GPT 5.6 "Sol"
• Mint-Ag
• Mint-Agent
• Mint-Cu
• Nex-N2 Mini
• RFC-Bench

• model release
• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →