Coinbase Says Fine-Tuned Qwen3.5-9B Outperforms Frontier AI Models in Fraud Detection

Coinbase $COIN CEO Brian Armstrong says the cryptocurrency exchange has fine-tuned the open-weight Qwen3.5-9B artificial intelligence model using internal company data, achieving stronger fraud detection performance than larger frontier models at a fraction of the cost.
The reported result highlights the potential of smaller, specialized AI systems to outperform more expensive general-purpose models on narrowly defined business tasks.
Fraud prevention is particularly important for cryptocurrency platforms, where account takeovers, suspicious transactions and social engineering attacks can create substantial financial and operational risks.
If the performance gains are reproducible, specialized open-weight models could help financial institutions reduce inference costs, improve detection workflows and maintain greater control over sensitive operational data.
However, the comparison is based on Coinbase's reported internal evaluation. The underlying benchmarks, cost savings and real-world false-positive rates have not been independently established in the supplied announcement.