The Model Zoo

Which open-weights models are worth the disk space right now, what they're good at, and what they hallucinate with confidence.

1. Our Model Stopped Saying No (2026-08-05)
2. You Can't Prompt Your Way to Both Commitment and Abstention (2026-08-11)
3. What Thinking Mode Actually Costs (2026-08-11)
4. Every Surprising Eval Result Was Our Infrastructure (2026-08-15)
5. A Safety Hook Truncated Every Code Generation (2026-08-15)
6. Every Failure Silently Scored 0.5 (2026-08-19)