Kamiwaza Agentic Performance Index

Benchmarks based on our PICARD agentic simulator framework. See how models perform in realistic enterprise-relevant tasks, no benchmaxxing thanks to PICARD's contamination-resistant design.

🗂️ Model Cards

Per model data at a glance.

View model cards →

Boards

🏆 Kamiwaza Agent Intelligence (KAI)

Overall agent intelligence score, combining KAMI and RIKER.

View index →

🏅 Kamiwaza Agentic Merit Index (KAMI) v1 Leaderboard

Our flagship agentic benchmark, using scenarios modeled against O*NET (Occupational Information Network) by the US Dept of Labor.

View leaderboard →

📚 RIKER v1 Leaderboard

Retrieval Intelligence and Knowledge Extraction Rating. Grounding, aggregation and fabrication at 128K, scored apart.

View leaderboard →

⚖️ Retrieval Frontier

Fidelity against cost. Which models nothing beats on both.

View frontier →

📈 Scaling and Peak Throughput

Throughput against parallel agents on 8xB300, FullBox mk5.

View scaling →

🧮 VRAM Planner

What fits on your box, and what it scores.

Open planner →

💠 Token Efficiency

Tokens spent per correct answer. Cheap and wrong is not cheap.

View efficiency →