OPEN LINK ↗
// 57m agoBENCHMARK RESULT
Kev 4B Quietly Reaches OpenRouter’s No. 4
Jared Palmer’s Kev 4B has reached fourth place on OpenRouter’s decision-model leaderboard despite minimal promotion. The open-weight model returns calibrated probabilities for typed yes/no, multiple-choice, and scoring questions without generating text.
// ANALYSIS
This is a strong adoption signal for compact, specialized models: developers are increasingly using dedicated decision layers instead of asking general-purpose LLMs to classify or route.
- –Built on Qwen3.5-4B with a LoRA adapter and pointer head
- –Supports routing, classification, triage, and policy checks through the System One-compatible API
- –Costs $0.042 per million input tokens with no output-token charge
- –Apache-2.0 weights make self-hosting and fine-tuning practical
- –OpenRouter rankings measure usage, not accuracy, so fourth place signals demand rather than universal model superiority
// TAGS
kev-4bllmopen-weightsinferencestructured-outputapievaluation
DISCOVERED
57m ago
2026-10-07
PUBLISHED
1h ago
2026-10-07
RELEVANCE
8/ 10
AUTHOR
phylera14