Hotep Intelligence Kush V9.1 Is Live: 100% Entity Grounding at 48.2 tok/s
Kush V9.1 clears every promotion gate that held back V9.0 — 7/7 Exact Entity Recall, 7/7 Prime Agent E2E, zero base-model bleed, and 48.2 tok/s on Metal.
Kush V9.1 clears every promotion gate that held back V9.0 — 7/7 Exact Entity Recall, 7/7 Prime Agent E2E, zero base-model bleed, and 48.2 tok/s on Metal.
Post-mortem on Kush V9: why Prime Agent E2E entity grounding gates triggered a production rollback to Kush V8.2 while V9.1 is hardened.
Kush V3 launches on Llama 3.1 8B with 103% more training data, chain-of-thought reasoning, and user feedback integration.
Inside Kush V2: 1,443 curated training pairs, zero data contamination, 184 entity-grounded examples, and validation across our 7-gate eval suite.
The first Hotep model on Meta's Llama 3.1 8B. A new naming convention, a new base architecture, and the bridge between v14's failures and Kush V2's triumph.