ZeroDrift launches three Anchor 3.0 models for real-time AI compliance checks

Startup ZeroDrift launched Anchor 3.0, a family of small language models that check AI agent messages against financial and corporate rules in real time. The flagship model caught over 90% of violations in a FINRA-based benchmark and, per the company, runs 100x faster and at under 1/500th the cost of GPT-6 Astra and Claude Fable 5.1 at comparable accuracy.

• Anchor 3.0 Mini: 9B-parameter MoE with 4B active parameters for high-volume traffic
• Flagship Anchor 3.0 applies 200+ FINRA and SEC rules and rewrites violations
• Anchor 3.0 Max: 27B parameters for long documents and internal policies
• Models post-trained from Gemma E4B and Qwen3.8-27B, available via Enforcement API