Wire · opportunities
Coinbase fine-tunes Qwen3.5-9B model for fraud prevention, beating frontier models
◆ Sectors
◆ Geography
◆ Source
◆ Verified
Fusion42 · 12 October 2026 · Fusion42 review
Coinbase fine-tuned Alibaba's Qwen3.5-9B AI model to outperform top frontier models in fraud detection on its Onramp crypto-buying service, reducing fraudulent transactions by 30% and training costs to under $100. The smaller model also ran twice as fast as competitors, signaling a strategic preference for specialised models over latest general-purpose AI systems.
This Wire brief sits within Fusion42's coverage of Fintech.
◆ ◆ The Wire takeaway
You need to rethink AI spend and model choice as small, cheap specialised models can now beat expensive frontier AI for fraud detection. Your risk tooling can get faster, cheaper, and more precise by switching focus from general-purpose giants to task-tuned models right now.
◆ Coverage
1 source · 11 Oct 2026
◆ Related on Wire
◆ Topics