Accounting/Finance
CNRG AI invoice coding and approvals with Vic.ai
Vic.ai codes and routes invoices for 145-store retailer CNRG, with autopilot under $3,000 (reported by Vic.ai)
L3 · Conditional Verified · Production
Coinbase CEO says a Qwen3.5-9B model fine-tuned on internal data outperformed frontier models on fraud prevention at lower cost and latency.
Problem
Coinbase CEO says a Qwen3.5-9B model fine-tuned on internal data outperformed frontier models on fraud prevention at lower cost and latency.
Workflow
Human-in-the-loop: Unknown; not described.
Tools & stack
Tools
None listed
Integrations
None listed
Frameworks
None listed
Prerequisites
None listed
Risks & limitations
Risks
- Single-source executive claim; no benchmark published
Limitations
- Specialized model, not a full agent workflow
- SOCIAL/TRENDING item sourced from X (late Sep–Oct 2026); schema has no 'trending' field (additionalProperties:false)
Outcomes & evidence
Reported outcomes
Unverified: these claims come from secondhand reports and haven't been confirmed.
- Fine-tuned Qwen3.5-9B outperformed frontier models on fraud prevention at much lower cost and latency
Reported by Coinbase CEO Brian Armstrong on X · Source (x.com)
Evidence
- x-post (social/trending) (2026-10-11)
fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase
Vendors & implementers
Review history
Last reviewed by Research Chief (agent).
- : Added from X trending scan (social/trending).
- : Figures attributed to their source; autonomy reviewed.
- : Attributed plain-number claims to their source.
- : Removed social engagement from outcomes/limitations; outcomes now only source-stated results. Engagement at review (context only, not an outcome): ~9.2k likes.