ChinaFinance and insurance2023 – ongoingLexinFintech Holdings
Singularity large model built on localized DeepSeek
LexinFintech said its in-house large model generated code 860,000 times a month and raised coding efficiency about 35%
Problem
LexinFintech, a Chinese consumer-finance company, runs high-volume customer service, telemarketing, collections and risk control, and has a large development team.
Approach
Lexin developed a proprietary large model called Singularity and deployed leading models such as DeepSeek locally, later upgrading to DeepSeek R1. Management said the model was used in daily operations including customer service, telemarketing, collections, coding, data analysis and risk control.
Results
Takeaway
These figures were stated by management on an earnings call. Lexin did not give the baseline for the 15% risk-identification gain beyond saying it was against the previous quarter.
Categories
- Industry
- Finance and insurance
- Business function
- Customer serviceFinance, risk and legalIT and R&D
- Innovation type
- Process innovation
- AI technology
- Generative AIPredictive analytics and ML
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
Transcript of Lexin's call of 18 March 2025. Management described the proprietary Singularity model and local deployment of DeepSeek, then R1. They gave the risk-identification, code-generation and coding-efficiency figures above.
Full details are in the original sources.