Coinbase Fine-Tuned AI Model to Detect Fraud
The company reduced fraudulent transaction volume by 30% using a specialized AI agent.
Updated on Oct. 11, 2026 in Artificial Intelligence
Live Poll
Do you trust specialized, smaller AI models to perform tasks better than general-purpose frontier models?
Coinbase has successfully fine-tuned the Qwen3.5-9B artificial intelligence model to identify fraudulent activity on its Onramp service. This lightweight model surpassed larger frontier models in speed while significantly cutting down on fraudulent transaction volume and total dollar losses.
Why it matters
By developing a high-performance model for under $100, Coinbase demonstrated that specialized, small-scale AI agents can be more efficient for specific security tasks than larger, general-purpose frontier models.
The fine-tuned Qwen model achieved a median latency of 0.683 seconds, which is 55% faster than the 1.515-second latency of the Opus 4.5 model. The training process for this agent utilized a single NVIDIA RTX PRO 6000 96GB GPU.
The players
Coinbase
Coinbase is a major American digital currency exchange platform that provides services for buying, selling, and storing cryptocurrency.
The details
The fraud detection agent operates by analyzing transaction data to provide a risk assessment layer atop existing machine-learning tools. Coinbase trained the model using reinforcement learning on a dataset comprising 16,140 total transactions, including 813 confirmed fraud cases.
Timeline
October 7, 2026: Coinbase released details regarding its fraud detection model in a blog series.
The Tech Race
The adoption of the Qwen3.5-9B model represents a broader trend in the tech industry where firms are shifting from relying on massive frontier models toward highly optimized, small-scale models. This approach positions Coinbase to achieve faster inference speeds and lower operational costs compared to competitors using traditional, larger AI architectures.
Users of the Coinbase Onramp service may experience faster transaction processing and more robust security against unauthorized account activity. Because the system is more efficient than previous models, the company can potentially maintain lower operational costs while strengthening consumer protection.
The takeaway
This case highlights how companies can achieve superior security outcomes by fine-tuning open-weight models for specific internal needs. Readers looking to improve operational efficiency should consider whether smaller, purpose-built AI agents can replace more expensive, generalized software solutions.
Further reading
For more information on the current landscape of AI development, explore our Artificial Intelligence section.
Live Poll
Do you trust specialized, smaller AI models to perform tasks better than general-purpose frontier models?
